The test is flaky, supposedly because of a bug in Lua JIT.
Disabling Lua JIT makes the test pass.

Workaround for #5081

I think that our current process is better. But if someone will want to
understand why we stick with it, it is good to have explicit cons and
pros.

This patch reworks approach to fiber management in Lua. Before
this patch each action that should return fiber led to new
userdata creation that was quite slow and made GC suffer. This
patch introduces new field in struct fiber to store a reference to
userdata that was created once for a fiber. It allows speedup
operations as fiber.self() and fiber.id().
Simple benchmark shows that access to fiber storage is faster in
two times, fiber.find() - 2-3 times and fiber.new/create functions
don't have any changes.

Before this patch we used an assumption in lua fiber code that
all valid lua refs are positive numbers. However there are
special constants defined in lua header for such purpose LUA_NOREF
and LUA_REFNIL [1]. This patch introduces special value in fiber
module that is equal to LUA_REFNIL to be sure that this value will
be correct in future static assert is added.

Also this patch introduces several usages of this value - let's
make fiber code more clearer and start to initialize ref values
with LUA_NOREF value. Also it will be needed for future changes.

  [1] https://pgl.yoyo.org/luai/i/luaL_ref

For future changes we will need the fiber object, not only its id.

Ref is a lua reference to fiber storage. This patch just renames
storage.lua.ref to storage.lua.storage_ref to underline that
it's not a reference to fiber itself and needed only for fiber
storage.

The problem is if we add a tuple to index and then fiber yield occurs (build-in-background),
replace of this tuple will change it in building index, but it will not be changed back
in case of rollback.

Closes #5958

Currently, if we delete tuple while primary index is altering in background,
a reference to removed tuple will be kept by primary index.

Closes #6290

Part of #5958

This patch fixes the comparison between DECIMAL as a left value and
DOUBLE greater than or equal to 1e38 or less than or equal to -1e38 as a
right value. Any DOUBLE value greater than or equal to 1e38 is now
greater than any DECIMAL value, and a DOUBLE value less than or equal to
-1e38 is less than any DECIMAL value. This is because our decimal cannot
contain more than 38 digits.

Closes #6376

Nullify all the output parameters passed to xrow_decode_subscribe().

This should help us get rid of cases when some of the parameters were
left uninitialized when the SUBSCRIBE request didn't contain the
corresponding fields.

Also it's now possible to remove all the explicit parameter
initializations for xrow_decode_* functions using
xrow_decode_subscribe() internally. Do that as well.

Follow-up #6034

Fixed dropping incoming messages when connection is closed or SHUT_RDWR
received and net_msg_max or readahead limit is reached. Now we don't close
connection if an error occurred while writing response, but only set a
special flag to stop further writing. Connection will be closed when we
read 0 from socket and process all requests.

Closes #6292

Currently there are several statistic metrics for iproto, which can
be obtained by using `box.stat.net`: `CONNECTIONS`, `REQUESTS`, `SENT`
and `RECEIVED`. Add new metric `STREAMS` which contain statistics for
iproto streams.

Part of #6293

@TarantoolBot document
Title: getting statistics for iproto streams was implemented
Add new  metric `STREAMS` to `box.stat.net`, which contain statistics
for iproto streams. `STREAMS` contains same counters as `CONNECTIONS`
metric in `box.stat.net`: current, rps and total. For example:
STREAMS:
  current: 0
  rps: 17980
  totat: 112623

Currently getting iproto statistics per thread is possible only
in debug mode, using `ERRINJ_IPROTO_SINGLE_THREAD_STAT` (it can be
used only in tests). This way have several disadvantages: first of
all `errinj` is not a place to get statistics, secondly user may be
interested in getting iproto statistics per thread. In this patch
`box.stat.net.thread` was implemented, it can be used for getting
same statistics as from `box.stat.net` but per thread.

Part of #6293

@TarantoolBot document
Title: getting network statistic per thread was implemented
User has possibility to run several iproto threads, but at the moment
he can get only general statistics, which combines statistics for all
threads. Now `box.stat.net.thread` has been implemented to get the same
statistics as when using `box.stat.net`, but for a thread. User can call
`box.stat.net.thread()` as a function to get general network statistics
per threads. For example for two iproto threads:
---
- - SENT:
      total: 0
      rps: 0
    CONNECTIONS:
      current: 0
      rps: 0
      total: 0
    REQUESTS:
      current: 0
      rps: 0
      total: 0
    RECEIVED:
      total: 0
      rps: 0
  - SENT:
      total: 0
      rps: 0
    CONNECTIONS:
      current: 0
      rps: 0
      total: 0
    REQUESTS:
      current: 0
      rps: 0
      total: 0
    RECEIVED:
      total: 0
      rps: 0
...

Also user can indexed it as a table by thread number. For example
`box.stat.net.thread[1]` returns network statistics for first iproto
thread:
---
- SENT:
    total: 0
    rps: 0
  CONNECTIONS:
    current: 0
    rps: 0
    total: 0
  REQUESTS:
    current: 0
    rps: 0
    total: 0
  RECEIVED:
    total: 0
    rps: 0
...

Make lbox_tuple_format_new (exported as box.internal.new_tuple_format to
Lua) reuse tuple formats. It's safe, because formats exported to Lua are
immutable. This fixes a net.box error when one creates/closes a lot of
net.box connections: a net.box connection creates a few formats for a
local version of the schema so if the garbage collector isn't invoked to
clean up after it, Tarantool can run out of format ids.

Closes #6217

An ephemeral space format doesn't have a dictionary so we ignore
dictionaries in format hashing and comparison functions. However,
formats exported to Lua (for net.box schema), which could also be
reused, do define field names. Since we'd like to reuse those as well,
we need to take field names into account while doing format hashing and
comparison.

Needed for #6217

Apart from ephemeral space formats, we can also reuse formats exported
to Lua, e.g. for net.box schema. Let's rename the flag and update the
comments to be relevant for all cases.

Needed for #6217

This patch adds xmalloc, xcalloc, xrealloc, xstrdup, and xstrndup helper
functions. Each of them calls the corresponding memory allocation
function and panics if it fails. See the issue description for the full
justification.

Closes #3534

This patch fixes the problem with copying an empty string in mem_copy().
Previously, because the string length was 0, an error was thrown, but
the diag was not set, which could lead to an error due to an empty diag
or to a double free.

Closes #6157
Closes #6399

It contains test case verifying that during parallel index build the
first index to finish wins the race and aborts another TX(s). For the
synchronization sake channels were used. However, in fact containing the
same number of tuples in indexes we can't predict which index finishes
first, so sometimes channel:get() executed inside fiber building index
may hang. Hence, let's simply fill one of spaces with a bit more tuples
and remove channel:get() from functions building indexes.

Closes #6354

In case we are updating "cluster" field in on_replace_dd_schema trigger
we do not check verifying that value corresponding to "cluster" key is
uuid. Meanwhile in contrast to other tuple_field_*() decoding functions
tuple_field_uuid() EXPECTS that passed value is type of mp_string. So
let's add extra string check to tuple_field_uuid().

Closes #6332

On replace data-dictionary triggers do not expect that exceptions can be
raised from their body. Unfortunately, on_replace_dd_schema still
contains one invocation of diag_raise() which may lead to unpredictable
results. So let's replace it with diag_set().

Part of #5332

In previous commit closing connection without dropping requests
was implemented. Now we can enable part of iproto transaction test
which depends on it. Also test was reworked to make it more clear
and all tests for 'vinyl' was removed because they are not related
to streams.

At the moment, when a net.box connection is closing we are cancel worker
fiber. In this case, data that was not sent is lost. In this patch connection
closing without dropping requests was implemented: before canceling fiber we
wait on conditional variable until all data will be sent.

Closes #6338

luaT_checkfibercond is necessary to implemenent closing connection
without unsent data lost. Also all other similar functions have a
different signiture.
PArt of #6338

To be able to configure libcurl for ARM64 CPU
on Fedora distros with linker hardening it is
necessary to enable `-fPIC`. It fails with wrong
relocation type for glibc symbols (e.g. socket)
otherwise.

Closes #6366

After this patch, DECIMAL values can be bound like any other supported
by SQL values.

Closes #4717

After this patch, getting an unknown msgpack extension in SQL will throw
the correct error.

Closes #6375

__tostring always returns the type name, which is 'net.box.request'.

__serialize returns user-defined fields attached to the future or an
empty table if there's none.

Closes #6314

The new version (2.10.0) is considered to be lower than 2.2.1, because
the test uses string comparison, which results in a test failure.

Remove the useless check.

Four changelog entries were fixed to resolve syntax errors.

The new release policy suggests to use X.Y.Z-alphaN, X.Y.Z-betaN,
X.Y.Z-rcN tags for pre-releases. See the policy document here:
https://github.com/tarantool/tarantool/discussions/6182

This commit enforces the new versioning for tarantool builds that're
made from CI.

Tarballs
--------

When we'll add the 2.10.0-beta1 tag, the tarball will be named so:

> tarantool-2.10.0-beta1.tar.gz

When we'll add one commit upward the tag, the tarball will be named so:

> tarantool-2.10.0-beta1.1.dev.tar.gz

(In fact, we don't deploy tarballs per push anymore, so it is just for
the same of completeness.)

For 2.10.0 release it'll be named as:

> tarantool-2.10.0.tar.gz

How to try: `make -f .gitlab.mk source`.

Packages
--------

When we'll add the 2.10.0-beta1 tag, the packages for Ubuntu Focal will
be named so:

> tarantool_2.10.0~beta1-1_amd64.deb
> tarantool-common_2.10.0~beta1-1_all.deb
> tarantool-dev_2.10.0~beta1-1_amd64.deb

The sources packaged together with the package will be named so:

> tarantool_2.10.0~beta1-1.debian.tar.xz
> tarantool_2.10.0~beta1-1.dsc
> tarantool_2.10.0~beta1.orig.tar.xz
> tarantool-2.10.0~beta1.tar.xz

When we'll add one commit upward the tag, those files will look so:

> tarantool_2.10.0~beta1.1.dev-1_amd64.deb
> tarantool-common_2.10.0~beta1.1.dev-1_all.deb
> tarantool-dev_2.10.0~beta1.1.dev-1_amd64.deb

> tarantool_2.10.0~beta1.1.dev-1.debian.tar.xz
> tarantool_2.10.0~beta1.1.dev-1.dsc
> tarantool_2.10.0~beta1.1.dev.orig.tar.xz
> tarantool-2.10.0~beta1.1.dev.tar.xz

(In fact, we don't deploy packages per push anymore, so it is just for
the same of completeness.)

For 2.10.0 release it'll be named so:

> tarantool_2.10.0.gcb3bdbf2a-1_amd64.deb
> tarantool-common_2.10.0.gcb3bdbf2a-1_all.deb
> tarantool-dev_2.10.0.gcb3bdbf2a-1_amd64.deb

> tarantool_2.10.0.gcb3bdbf2a-1.debian.tar.xz
> tarantool_2.10.0.gcb3bdbf2a-1.dsc
> tarantool_2.10.0.gcb3bdbf2a.orig.tar.xz
> tarantool-2.10.0.tar.xz

How to try: `OS=ubuntu DIST=focal make -f .gitlab.mk package`.

Fixes #6184

See the previous commit 'test: descrease instance file name length' for
the problem description.

Since we're going to use the tags like 2.10.0-beta1 for upcoming
pre-releases, we have even less symbols for a filename of a unix socket.
So I chose the shorter names on the test-run side: foo.socket-iproto ->
foo.i, foo.socket-admin -> foo.c (console).

See https://github.com/tarantool/test-run/pull/312

This update is mostly necessary for the current master branch, but it'll
be applied for 1.10 as well: we keep test-run consistent across alive
branches.

Part of #6184

First, several facts:

* The maximal unix socket file length is 108 symbols on Linux (in fact,
  107 for tarantool console socket, see #4634).
* We run testing during building of RPM packages.
* When tests are run, they use the build directory for temporary testing
  files and tarantool console unix socket files.
* Test-run uses an instance script name to form a unix socket name for
  tarantool console.
* We plan to change package version format and add .dev suffix for
  transient (non tagged) commits.
* The package version participates in the build directory name, when
  we're building an RPM package.

If we'll enable the new version format just now, we'll unable to listen
on the unix socket like the following (the length is 111):

```
/build/usr/src/debug/tarantool-2.9.0.353.dev/test/var/023_box/gh-3633-simple-tuple-size-increasing.socket-admin
```

So I changed the instance file name to make it shorter.

Of course, there are ways to solve the problem at all, this commit is
just workaround.

Part of #6184
Follows up #3633

Was added 'export DEBIAN_FRONTEND=noninteractive', because there are
packages, which are noninteractive by default. So, CI raises
'Unable to acquire the dpkg frontend lock (/var/lib/dpkg/lock-frontend), is another process using it?'.

Were added workflows with *.rpm or *.deb packaging for arm architecture.
Those workflows only verify package builds and don't deploy them to our
repositories for now.

Part of #6222

Tests were turned off from arm workflows according to instability.

Needed for #6222

It requires update_repo.sh changes, which are not trivial in
verification. We want to be sure that all apt/yum metainformation is
generated correctly, so we'll postpone the automatic deployment task.

Part of #6222

Follows up #6185