Credentials is a cache of user universal privileges. And that
cache can become outdated in case user privs were changed after
creation of the cache.

The patch makes user update all its credentials caches with new
privileges, via a list of all creds.

That solves a couple of real life problems:

- If a user managed to connect after box.cfg started listening
port, but before access was granted, then he needed a reconnect;

- Even if access was granted, a user may connect after box.cfg
listen, but before access *is recovered* from _priv space. It
was not possible to fix without a reconnect. And this problem
affected replication.

Closes #2763
Part of #4535
Part of #4536

@TarantoolBot document
Title: User privileges update affects existing sessions and objects
Previously if user privileges were updated (via
`box.schema.user.grant/revoke`), it was not reflected in already
existing sessions and objects like functions. Now it is.

For example:
```
        box.cfg{listen = 3313}
        box.schema.user.create('test_user', {password = '1'})
        function test1() return 'success' end

        c = require('net.box').connect(box.cfg.listen, {
                user = 'test_user', password = '1'
        })
        -- Error, no access for this connection.
        c:call('test1')

        box.schema.user.grant('test_user', 'execute', 'universe')
        -- Now works, even though access was granted after
        -- connection.
        c:call('test1')
```

A similar thing happens now with `box.session.su` and functions
created via `box.schema.func.create` with `setuid` flag.

In other words, now user privileges update is reflected
everywhere immediately.

(cherry picked from commit 06dbcec597f14fae6b3a7fa2361f2ac513099662)
(cherry picked from commit 2b599c0efa9ae265fb7464af6abae3f6a192e30e)

End recovery (which means building secondary indexes) just after
last known log file was read. This allows fast switch to hot standby
instance without any delay for secondary index to be built.
Due to engine_end_recovery carryover, xdir_collect_inprogress,
previously being called from it, is now moved to garbage collector.

Closes #4135

(cherry picked from commit 5aa243de)

Rows_per_wal option was deprecated because it can be covered by
wal_max_size. In order not to complicate WAL code with that
option's support this commit drops it completely.

In some tests the option was used to create several small xlog
files. Now the same is done via wal_max_size. Where it was
needed, number of rows per wal is estimated as wal_max_size / 50.
Because struct xrow_header size ~= 50 not counting paddings and
body.

Note, file box/configuration.result was deleted here, because it
is a stray result file, and it contained the rows_per_wal option
mentioning. Its test was dropped much earlier in
fdc3d1dd.

Closes #3762

(cherry picked from commit c6012920)

Closes #4434
Follow-up #4366

@TarantoolBot document
Title: json/msgpack.cfg.encode_deep_as_nil option

Tarantool has several so called serializers to convert data
between Lua and another format: YAML, JSON, msgpack.

YAML is a crazy serializer without depth restrictions. But for
JSON, msgpack, and msgpackffi a user could set encode_max_depth
option. That option led to crop of a table when it had too many
nested levels. Sometimes such behaviour is undesirable.

Now an error is raised instead of data corruption:

    t = nil
    for i = 1, 100 do t = {t} end
    msgpack.encode(t) -- Here an exception is thrown.

To disable it and return the old behaviour back here is a new
option:

    <serializer>.cfg({encode_deep_as_nil = true})

Option encode_deep_as_nil works for JSON, msgpack, and msgpackffi
modules, and is false by default. It means, that now if some
existing users have cropping, even intentional, they will get the
exception.

(cherry picked from commit d7a8942a)

Tuple is a C library exposed to Lua. In Lua to translate Lua
objects into tuples and back luaL_serializer structure is used.

In Tarantool we have several global serializers, one of which is
for msgpack. Tuples store data in msgpack, and in theory should
have used that global msgpack serializer. But in fact the tuple
module had its own private serializer because of tuples encoding
specifics such as never encode sparse arrays as maps.

This patch makes tuple Lua module use global msgpack serializer
always. But how does tuple handle sparse arrays now? In fact,
the tuple module still has its own serializer, but it is updated
each time when the msgpack serializer is changed.

Part of #4434

(cherry picked from commit 676369b1)

In a configuration with several read-only and read-write instances, if
replication_connect_quorum is not greater than the amount of read-only
instances and replication_connect_timeout happens to be small enough
for some read-only instances to form a quorum and exceed the timeout
before any of the read-write instaces start, all these read-only
instances will choose themselves a read-only bootstrap leader.
This 'leader' will successfully bootstrap itself, but will fail to
register any of the other instances in _cluster table, since it isn't
writeable. As a result, some of the read-only instances will just die
unable to bootstrap from a read-only bootstrap leader, and when the
read-write instances are finally up, they'll see a single read-only
instance which managed to bootstrap itself and now gets a
REPLICASET_UUID_MISMATCH error, since no read-write instance will
choose it as bootstrap leader, and will rather bootstrap from one of
its read-write mates.

The described situation is clearly not what user has hoped for, so
throw an error, when a read-only instance tries to initiate the
bootstrap. The error will give the user a cue that he should increase
replication_connect_timeout.

Closes #4321

@TarantoolBot document
Title: replication: forbid to bootstrap read-only masters.

It is no longer possible to bootstrap a read-only instance in an emply
data directory as a master. You will see the following error trying to
do so:
```
ER_BOOTSTRAP_READONLY: Trying to bootstrap a local read-only instance as master
```
Now if you have a fresh instance, which has
`read_only=true` in an initial `box.cfg` call, you need to set up
replication from an instance which is either read-write, or has your
local instance's uuid in its `_cluster` table.

In case you have multiple read-only and read-write instances with
replication set up, and you still see the aforementioned error message,
this means that none of your read-write instances managed to start
listening on their port before read_only instances have exceeded the
`replication_connect_timeout`. In this case you should raise
`replication_connect_timeout` to a greater value.

(cherry picked from commit 037bd58c)

Added "fragile" option to the flaky tests that are
not intended to be run in parallel with others.
Option set at the suite.ini file at the appropriate
suites with comments including the issue that stores
the fail.

(cherry picked from commit 165f8ee6)

It turned out that patch d0e38d59 wasn't accompanied
w/ a test. Add proper check.

(cherry picked from commit c80e9416)

A txn_stmt keeps a reference to the space it modifies. Memtx uses this
space reference to revert the statement on error or voluntary rollback
so the space must stay valid throughout the whole transaction.

The problem is a DML statement may be followed by a DDL statement that
modifies the target space in the same transaction. If we try to roll
it back before running the rollback triggers installed by the DDL
statement, it will access an invalid space object (e.g. missing an
index), which will result in a crash.

To fix this problem, let's run triggers installed by a statement right
after rolling back the statement.

Closes #4368

After 2.2.0-390-ga7c855e5b ("net.box: fetch '_vcollation' sysview into
the module") net.box fetches _vcollation view unconditionally, while the
view was added in 2.2.0-389-g3e3ef182f and, say, tarantool-1.10 and
tarantool-2.1 do not have it. This leads to a runtime error "Space '277'
does not exist" on a newer client that connects to an older server.

Now the view is fetched conditionally depending of a version of a
server: if it is above 2.2.1, then net.box will fetch it. Note: at the
time there are no release with a number above 2.2.1.

When _vcollation view is available, a collation in an index part will be
shown by its name (with 'collation' field), otherwise it will be shown
by its ID (in 'collation_id' field). For example:

Connect to tarantool 1.10:

 | tarantool> connection = require('net.box').connect('localhost:3301')
 | ---
 | ...
 |
 | tarantool> connection.space.s.index.sk.parts
 | ---
 | - - type: string
 |     is_nullable: false
 |     collation_id: 2
 |     fieldno: 2
 | ...

Connect to tarantool 2.2.1 (when it will be released):

 | tarantool> connection = require('net.box').connect('localhost:3301')
 | ---
 | ...
 |
 | tarantool> connection.space.s.index.sk.parts
 | ---
 | - - type: string
 |     is_nullable: false
 |     collation: unicode_ci
 |     fieldno: 2
 | ...

Fixes #4307.

When reverting to a savepoint inside a DDL transaction, apart from
undoing changes done by the DDL statements to the system spaces, we also
have to

 - Run rollback triggers installed after the savepoint was set, because
   otherwise changes done to the schema by DDL won't be undone.
 - Remove commit triggers installed after the savepoint, because they
   are not relevant anymore, apparently.

To achieve that let's append DDL triggers right to txn statements.
This allows us to easily discard commit triggers and run rollback
triggers when a statement is rolled back.

Note, txn commit/rollback triggers are not removed, because they are
still used by applier and Lua box.on_commit/on_rollback functions.

Closes #4364
Closes #4365

Closes #1260

@TarantoolBot document
Title: introduce func indexes in memtx
Now you can define a func_index using a registered persistent
function.

There are restrictions for function and key definition for
a functional index:
 - the referenced function must be persistent, deterministic
   and must return a scalar type or an array.
 - you must define key parts which describe the function return value
 - the function must return data which types match the
   defined key parts
 - the function may return multiple keys; this would be a multikey
   functional index; each key entry is indexed separately;
 - for multikey functional indexes, the key definition should
   start with part 1 and cover all returned key parts
 - key parts can't use JSON paths.
 - the function used for the functional index can not access tuple
   fields by name, only by index.

Functional index can't be primary.
It is not possible to change the used function after a functional
index is defined on it. The index must be dropped first.

Each key returned by functional index function (even when it is a
single scalar) must be returned as a table i.e. {1} and must
match the key definition.

To define a multikey functional index, create a function with
opts = {is_multikey = true} and return a table of keys.

Example:
s = box.schema.space.create('withdata')
s:format({{name = 'name', type = 'string'},
          {name = 'address', type = 'string'}})
pk = s:create_index('name', {parts = {1, 'string'}})
lua_code = [[function(tuple)
                local address = string.split(tuple[2])
                local ret = {}
                for _, v in pairs(address) do
			table.insert(ret, {utf8.upper(v)})
		end
                return ret
             end]]
box.schema.func.create('address', {body = lua_code,
                       is_deterministic = true, is_sandboxed = true,
                       opts = {is_multikey = true}})
idx = s:create_index('addr', {unique = false,
                     func = 'address',
                     parts = {{1, 'string', collation = 'unicode_ci'}}})
s:insert({"James", "SIS Building Lambeth London UK"})
s:insert({"Sherlock", "221B Baker St Marylebone London NW1 6XE UK"})
idx:select('Uk')
---
- - ['James', 'SIS Building Lambeth London UK']
  - ['Sherlock', '221B Baker St Marylebone London NW1 6XE UK']
...

Needed for #1260

@TarantoolBot document
Title: A new option is_multikey for function definition

A new option is_multikey allows to specify wether new function
returns multiple values packed in a table object. This is a
native way to define multikey func_index.

@TarantoolBot document
Title: document \set output lua

Historically we use YAML format to print results of operation to
a console. Moreover our test engine is aiming YAML as a primary format
to compare results of test runs. Still we need an ability to print
results in a different fasion, in particular one may need to use
the console in a REPL way so that the results would be copied and
pased back to further processing.

For this sake we introduce that named "output" command which allows
to specify which exactly output format to use. Currently only yaml
and lua formats are supported.

To specify lua output format type

 | tarantool> \set output lua

in the console. lua mode supports line oriented output (default) or
block mode.

For example

 | tarantool> a={1,2,3}
 | tarantool> a
 | ---
 | - - 1
 |   - 2
 |   - 3
 | ...
 | tarantool> \set output lua
 | true
 | tarantool> a
 | {1, 2, 3}
 | tarantool> \set output lua,block
 | true
 | tarantool> a
 | {
 |   1,
 |   2,
 |   3
 | }

By default YAML output format is kept for now, simply to not
break the test engine. The output is bound to a session, thus every
new session should setup own conversion if needed.

Since serializing lua data is not a trivial task we use "serpent"
third party module to convert data.

Part-of #3834

In contrast to subsequent calls, the initial call to box.cfg didn't log
configuration changes to the default state. As a result, by looking at
a log file we coudn't tell which configuration was being used.

Closes #4236

This last error
```
[035]  ...
[035]  disconnected_cnt
[035]  ---
[035] -- 1
[035] +- 2
[035]  ...
[035]  conn:close()
[035]  ---
[035]  ...
[035]  disconnected_cnt
[035]  ---
[035] -- 2
[035] +- 3
[035]  ...
[035]  test_run:cmd('stop server connecter')
[035]  ---
[035]
```
Happens because net.box is able to connect to tarantool before it has
finished bootstrap. When connecting, net.box tries to fetch schema
executing a couple of selects, but fails to pass access check since
grants aren't applied yet. This is described in detail in
https://github.com/tarantool/tarantool/issues/2763#issuecomment-499046998
So, alter the test so that it tolerates multiple connection failures.

Closes #4273

The "box/net.box.test.lua" test contains a check that the error
received contains a 'timed out'. But in cases when testing was
conducted on a slow computer or in the case of a very large load,
it is possible that the connection time-out will be reached
earlier than the mentioned error. In this case, the error "Invalid
argument" will be returned. To prevent this from happening,
this patch will increase the connection timeout.

Closes #4341

The patch is pretty straightforward - all it does is moves checks for
single statement transactions from alter.cc to txn_enable_yield_for_ddl
so that now any DDL request may be executed in a transaction unless it
builds an index or checks the format of a non-empty space (those are the
only two operations that may yield).

There's two things that must be noted explicitly. The first is removal
of an assertion from priv_grant. The assertion ensured that a revoked
privilege was in the cache. The problem is the cache is built from the
contents of the space, see user_reload_privs. On rollback, we first
revert the content of the space to the original state, and only then
start invoking rollback triggers, which call priv_grant. As a result, we
will revert the cache to the original state right after the first
trigger is invoked and the following triggers will have no effect on it.
Thus we have to remove this assertion.

The second subtlety lays in vinyl_index_commit_modify. Before the commit
we assumed that if statement lsn is <= vy_lsm::commit_lsn, then it must
be local recovery from WAL. Now it's not true, because there may be
several operations for the same index in a transaction, and they all
will receive the same signature in on_commit trigger. We could, of
course, try to assign different signatures to them, but that would look
cumbersome - better simply allow lsn <= vy_lsm::commit_lsn after local
recovery, there's actually nothing wrong about that.

Closes #4083

@TarantoolBot document
Title: Transactional DDL

Now it's possible to group non-yielding DDL statements into
transactions, e.g.

```Lua
box.begin()
box.schema.space.create('my_space')
box.space.my_space:create_index('primary')
box.commit() -- or box.rollback()
```

Most DDL statements don't yield and hence can be run from transactions.
There are just two exceptions: creation of a new index and changing the
format of a non-empty space. Those are long operations that may yield
so as not to block the event loop for too long. Those statements can't
be executed from transactions (to be more exact, such a statement must
go first in any transaction).

Also, just like in case of DML transactions in memtx, it's forbidden to
explicitly yield in a DDL transaction by calling fiber.sleep or any
other yielding function. If this happens, the transaction will be
aborted and an attempt to commit it will fail.

Memtx engine doesn't allow yielding inside a transaction. To achieve
that, it installs fiber->on_yield trigger that aborts the current
transaction (rolls it back, but leaves it be so that commit fails).

There's an exception though - DDL statements are allowed to yield.
This is required so as not to block the event loop while a new index
is built or a space format is checked. Currently, we handle this
exception by checking space id and omitting installation of the
trigger for system spaces. This isn't entirely correct, because we
may yield after a DDL statement is complete, in which case the
transaction won't be aborted though it should:

  box.begin()
  box.space.my_space:create_index('my_index')
  fiber.sleep(0) -- doesn't abort the transaction!

This patch fixes the problem by making the memtx engine install the
on_yield trigger unconditionally, for all kinds of transactions, and
instead explicitly disabling the trigger for yielding DDL operations.

In order not to spread the yield-in-transaction logic between memtx
and txn code, let's move all fiber_on_yield related stuff to txn,
export a method to disable yields, and use the method in memtx.

Closes #4182
Closes #4219
Needed for #1260

@TarantoolBot document
Title: Persistent Lua functions

Now Tarantool supports 'persistent' Lua functions.
Such functions are stored in snapshot and are available after
restart.
To create a persistent Lua function, specify a function body
in box.schema.func.create call:
e.g. body = "function(a, b) return a + b end"

A Lua persistent function may be 'sandboxed'. The 'sandboxed'
function is executed in isolated environment:
  a. only limited set of Lua functions and modules are available:
    -assert -error -pairs -ipairs -next -pcall -xpcall -type
    -print -select -string -tonumber -tostring -unpack -math -utf8;
  b. global variables are forbidden

Finally, the new 'is_deterministic' flag allows to mark a
registered function as deterministic, i.e. the function that
can produce only one result for a given list of parameters.

The new box.schema.func.create interface is:
box.schema.func.create('funcname', <setuid = true|FALSE>,
	<if_not_exists = true|FALSE>, <language = LUA|c>,
	<body = string ('')>, <is_deterministic = true|FALSE>,
	<is_sandboxed = true|FALSE>, <comment = string ('')>)

This schema change is also reserves names for sql builtin
functions:
    TRIM, TYPEOF, PRINTF, UNICODE, CHAR, HEX, VERSION,
    QUOTE, REPLACE, SUBSTR, GROUP_CONCAT, JULIANDAY, DATE,
    TIME, DATETIME, STRFTIME, CURRENT_TIME, CURRENT_TIMESTAMP,
    CURRENT_DATE, LENGTH, POSITION, ROUND, UPPER, LOWER,
    IFNULL, RANDOM, CEIL, CEILING, CHARACTER_LENGTH,
    CHAR_LENGTH, FLOOR, MOD, OCTET_LENGTH, ROW_COUNT, COUNT,
    LIKE, ABS, EXP, LN, POWER, SQRT, SUM, TOTAL, AVG,
    RANDOMBLOB, NULLIF, ZEROBLOB, MIN, MAX, COALESCE, EVERY,
    EXISTS, EXTRACT, SOME, GREATER, LESSER, SOUNDEX,
    LIKELIHOOD, LIKELY, UNLIKELY,
    _sql_stat_get, _sql_stat_push, _sql_stat_init, LUA

A new Lua persistent function LUA is introduced to evaluate
LUA strings from SQL in future.

This names could not be used for user-defined functions.

Example:
lua_code = [[function(a, b) return a + b end]]
box.schema.func.create('summarize', {body = lua_code,
		is_deterministic = true, is_sandboxed = true})
box.func.summarize
---
- aggregate: none
  returns: any
  exports:
    lua: true
    sql: false
  id: 60
  is_sandboxed: true
  setuid: false
  is_deterministic: true
  body: function(a, b) return a + b end
  name: summarize
  language: LUA
...
box.func.summarize:call({1, 3})
---
- 4
...

@kostja: fix style, remove unnecessary module dependencies,
add comments

Test box/net.box.test.lua checks state of the connection in case
of an error. It should be 'error_reconnect'. But, in cases where
testing was performed on a slow computer or in the case of a very
large load, it is possible that the connection status may change
from the 'error_reconnect' state to another state. This led to the
failure of the test. Since this check is not the main purpose of
the test, it is better to simply delete the check.

Closes #4335

Quoting feature request

 | Tarantool is Database and Application Server in one box.
 |
 | Appserver development process contains a lot of
 | lua/luajit-ffi/lua-c-extension code.
 |
 | Coredump is very useful in case when some part of appserver crashed.
 | If the reason is input - data from database is not necessary. If the reason
 | is output - data from database is already in snap/xlog files.
 |
 | Therefore consider core dumps without data enabled by default.

For info: the strip_core feature has been introduced in
549140b3

Closes #4337

@TarantoolBot document
Title: Document box.cfg.strip_core

When Tarantool runs under a heavy load the memory allocated
for tuples may be very huge in size and to eliminate this
memory from being present in `coredump` file the `box.cfg.strip_core`
parameter should be set to `true`.

The default value is `true`.

box/net.box test flaky failed on grepping the log file
for 'ER_NO_SUCH_PROC' pattern on high load running hosts,
found that the issue can be resolved by updating the
grep_log to wait_log function to make able to wait the
needed message for some time.

[008] Test failed! Result content mismatch:
[008] --- box/net.box.result	Tue Jul  9 17:00:24 2019
[008] +++ box/net.box.reject	Tue Jul  9 17:03:34 2019
[008] @@ -1376,7 +1376,7 @@
[008]  ...
[008]  test_run:grep_log("default", "ER_NO_SUCH_PROC")
[008]  ---
[008] -- ER_NO_SUCH_PROC
[008] +- null
[008]  ...
[008]  box.schema.user.revoke('guest', 'execute', 'universe')
[008]  ---

Closes #4329

There is compile time option PACKAGE in cmake to define
current build distribution info. For community edition
is it Tarantool by default. For enterprise it is
Tarantool Enterprise

There were no option to check distribution name in runtime.
This change adds box.info.package output for CE and TE.

The test regarding logging corrupted rows failed occasionally with
```
[016]  test_run:grep_log('default', 'Got a corrupted row.*')
[016]  ---
[016] -- 'Got a corrupted row:'
[016] +- null
[016]  ...
```
The logs then had
```
[010] 2019-07-06 19:36:16.857 [13046] iproto sio.c:261 !> SystemError writev(1),
called on fd 23, aka unix/:(socket), peer of unix/:(socket): Broken pipe
```
instead of the expected message.

This happened, because we closed a socket before tarantool could write a
greeting to the client, the connection was then closed, and execution
never got to processing the malformed request and thus printing the
desired message to the log.

To fix this, actually read the greeting prior to writing new data and
closing the socket.

Follow-up #4273

When a memtx transaction is aborted on yield, it isn't enough to
rollback individual statements - we must also run on_rollback triggers,
otherwise changes done to the schema by an aborted DDL transaction will
be visible to other fibers until an attempt to commit it is made.

The test case has two problems that appear from time to time and lead to
flaky fails. Those fails are look as shown below in a test-run output.

 | Test failed! Result content mismatch:
 | --- box/net.box.result	Mon Jun 24 17:23:49 2019
 | +++ box/net.box.reject	Mon Jun 24 17:51:52 2019
 | @@ -1404,7 +1404,7 @@
 |  ...
 |  test_run:grep_log('default', 'ER_INVALID_MSGPACK.*')
 | ---
 | -- 'ER_INVALID_MSGPACK: Invalid MsgPack - packet body'
 | +- 'ER_INVALID_MSGPACK: Invalid MsgPack - packet length'
 | ...
 | -- gh-983 selecting a lot of data crashes the server or hangs the
 | -- connection

'ER_INVALID_MSGPACK.*' regexp should match 'ER_INVALID_MSGPACK: Invalid
MsgPack - packet body' log message, but if it is not in a log file at a
time of grep_log() call (just don't flushed to the file yet) a message
produced by another test case can be matched ('ER_INVALID_MSGPACK:
Invalid MsgPack - packet length'). The fix here is to match the entire
message and check for the message periodically during several seconds
(use wait_log() instead of grep_log()).

Another problem is the race between writing a response to an iproto
socket on a server side and closing the socket on a client end. If
tarantool is unable to write a response, it does not produce the warning
re invalid msgpack, but shows 'broken pipe' message instead. We need
first grep for the message in logs and only then close the socket on a
client. The similar problem (with another test case) is described in
[1].

[1]: https://github.com/tarantool/tarantool/issues/4273#issuecomment-508939695

Closes: #4311

There were tests relying on certain content of SWIM messages.
After next patches these conditions won't work without an
explicit intervention with error injections.

The patchset moves these tests to separate release-disabled
files.

Part of #4253

This patch adds a stack cleanup after a trigger is run and its return
values, if any, have been read.

This problem was found in a case when on_schema_init trigger set an
on_replace trigger on a space, and the trigger ran during recovery.
This lead to Lua stack overflows for the aforementioned reasons.

Closes #4275

Now, as we don't need to take the schema lock for checkpointing, it is
only used to synchronize concurrent space modifications (drop, truncate,
alter). Actually, a global lock is a way too heavy means to achieve this
goal, because we only care about forbidding concurrent modifications of
the same space while concurrent modifications of different spaces should
work just fine. So this patch replaces the global schema lock with per
space locking.

A space lock is held while alter_space_do() is in progress so as to make
sure that while AlterSpaceOp::prepare() is performing a potentially
yielding operation, such as building a new index, the space struct
doesn't get freed from under our feet. Note, the lock is released right
after index build is complete, before the transaction is committed to
WAL, so if the transaction is non-yielding it can modify the space again
in the next statement (this is impossible now, but will be done in the
scope of the transactional DDL feature).

If alter_space_do() sees that the space is already locked it bails out
and throws an error. This should be fine, because long-lasting operation
involving schema change, such as building an index, are rare and only
performed under the supervision of the user so throwing an error rather
than waiting seems to be adequate.

Removal of the schema lock allows us to remove latch_steal() helper and
on_begin_stmt txn trigger altogether, as they were introduced solely to
support locking.

This is a prerequisite for transactional DDL, because it's unclear how
to preserve the global schema lock while allowing to combine several DDL
statements in the same transaction.

Implemented GitLab CI testing process additionally to existing Travis
CI. The new testing process is added to run tests faster. It requires to
control a load of machines to avoid flaky fails on timeouts. GitLab CI
allows us to run testing on our machines.

Created 2 stages for testing and deploying packages.

The testing stage contains the following jobs that are run for all
branches:

* Debian 9 (Stretch): release/debug gcc.
* Debian 10 (Buster): release clang8 + lto.
* OSX 14 (Mojave): release.
* FreeBSD 12: release gcc.

And the following jobs that are run of long-term branches (release
branches: for now it is 1.10, 2.1 and master):

* OSX 13 (Sierra): release clang.
* OSX 14 (Mojave): release clang + lto.

The deployment stage contains the same jobs as we have in Travis CI.
They however just build tarballs and packages: don't push them to S3 and
packagecloud.

In order to run full testing on a short-term branch one can name it with
'-full-ci' suffix.

The additional manual work is needed when dependencies are changed in
.travis.mk file ('deps_debian' or 'deps_buster_clang_8' goals):

 | make GITLAB_USER=foo -f .gitlab.mk docker_bootstrap

This command pushes docker images into GitLab Registry and then they are
used in testing. Pre-built images speed up testing.

Fixes #4156

Timeout injections are unstable and difficult to use. Injecting a delay
is much more convenient.

A sequence isn't supposed to roll back to the old value if the
transaction it was used in is aborted for some reason. However,
if a sequence is dropped, we do want to restore the original
value on rollback so that we don't lose it on an unsuccessful
attempt to drop the sequence.

_space_sequence changes are not rolled back properly. Fix it keeping in
mind that our ultimate goal is to implement transactional DDL, which
implies that all changes to the schema should be done synchronously,
i.e. on_replace, not on_commit.

To implement transactional DDL, we must make sure that in-memory schema
is updated synchronously with system space updates, i.e. on_replace, not
on_commit.

Note, to do this in case of the sequence cache, we have to rework the
way sequences are exported to Lua - make on_alter_sequence similar to
how on_alter_space and on_alter_func triggers are implemented.

To implement transactional DDL, we must make sure that in-memory schema
is updated synchronously with system space updates, i.e. on_replace, not
on_commit.

Replace prints that indicate on_shutdown trigger execution with
log.warn, which is more reliable. This eliminates occasional test
failures. Also instead of waiting for the server to start and executing
grep_log, wait for the desired log entries to appear with wait_log.

Closes #4134

A new VARBINARY field type would be useful for SQL type system.

Closes #4201
Needed for #4206

@TarantoolBot document
Title: new varbinary field type

Introduced a new field type varbinary to represent mp_bin values.
The new type varbinary may be used in format or index definition.

Example:
s = box.schema.space.create('withdata')
s:format({{"b", "varbinary"}})
pk = s:create_index('pk', {parts = {1, "varbinary"}})

Fixes commit abf5ef4f ("box: export registered functions in box.func
folder").

In SQL, it is allowed to use vector expressions, that is, an
operation that uses vectors as operands. For instance, vector
comparison:
SELECT (1,2,3) < (1,2,4);

Accidentally, routines handling IN operator contained a bug: in
cases where we used a vector as the left value in the IN operator,
we received an assertion in debug build or a segmentation fault in
release. This was due to some legacy code in which it was assumed
that the left value of the IN operator can have only one column in
case it is a vector. Let's fix this by allowing vectors of the
other sizes as the left value of the IN operator and providing
check which verifies that both sides of IN operator have the same
dimension.

Closes #4204