Current func_adapter implementation has several drawbacks. Firstly, the
interface has 21 virtual method (except for virtual destructor) which
means you need to implement all these methods if you need a new
func_adapter. Secondly, if you need to pass arguments from Lua to
func_adapter, you need to iterate over Lua stack and push the value to
func_adapter with appropriate type.

We have internal interface for calling persistent functions - port.
Let's use it for triggers too. It will allow to easily implement new
func_adapter since it will have only call method and virtual destructor.
Also, it will allow to easily call any func_adapter from Lua using
port_lua - now module trigger relies on the fact that there are only Lua
triggers. This change reduces code complexity as well - since we are
using port_c for triggers, that allows to be dumped several times, we
can create helper `event_run_triggers` which incapsulates work with
`event_trigger_iterator` and simply calls all the triggers from the
event passing the same port with arguments.

Since func_adapter is used mostly for triggers, let's make both ports
arguments optional - many triggers have no arguments and almost all
triggers ignore returned values.

Along the way, fix a possible crash when iterator passed to
transactional trigger is saved and used after the trigger is over.

NO_CHANGELOG=internal
NO_DOC=internal

The method allows to represent port contents as a list of port_c_entry.
It is handy to easily iterate over the port contents since port_c_entry
is actually a list of variants of supported by Tarantool types. It is
needed to easily process values returned from triggers.

Returned entries mustn't be destroyed since they don't own any resources.
Their lifetime is similar to the MsgPack obtained by port_get_msgpack -
it may be returned directly from the port, or it can be allocated on
fiber()->gc, so the caller is responsible for cleaning the region up.

NO_CHANGELOG=internal
NO_DOC=internal

We are going to use port_c for user-defined triggers, so we need to add
an ability to store iterator in port_c. Also, this port already allows
to be dumped several times - we should keep this invariant to call
chains of triggers using only one port.

Therefore, this commit populates port_c with new type called `iterable`.
An iterable object is an object that can create iterators - an iterator
is created on each dump. We decided that such approach is more
convenient than copying an iterator in initial state on each dump.

Iterable object and created iterator have no destructor and any
invalidation logic, so author of pushed object must think about iterator
invalidation by himself. What's about destructors, user of port_c must
ensure that the object is alive as long as the port can be dumped. When
port_c is not needed anymore, the object can be deleted by caller.
Created iterator has no destructor because we think that iterator
shouldn't own any resources - its state is allocated when the iterable
object is dumped, and the release of the state lies on the side that dumped
the object. Rationale: if you dump iterable object to Lua, the
iterator is most likely to be freed only when Lua VM will run out of
memory (GC will be called then), so Tarantool resources will be occupied
for a long time. Let's try to avoid such situation.

NO_CHANGELOG=internal
NO_DOC=internal

We are going to use ports for triggers. The problem is there is no port
implementation that meets all requirements to be used to pass arguments
to triggers from C, so let's extend existing port_c. This port is
already allows to be dumped several times - let's keep this invariant to
call chains of triggers using only one port.

Let's store trivial types directly, not packed in  MsgPack. Then, we won't
need to make a second mempool allocation to store them and we won't need
to decode them to read.

It's worth noting that port_c has two MsgPack types now - MP and
MP_OBJECT. The first now is a MsgPack packet that will be unpacked on
dump if it's possible. For example, we can pass a field of tuple, which
is a MsgPack packet, to field constraint, and it will be unpacked when
it will be dumped to Lua, so the constraint argument will have a Lua
value, not MsgPack. On the other hand, we want to dump MsgPack as a
MsgPack object sometimes. For instance, we use MsgPack object to pass
request header and body to iproto override handlers and space recovery
triggers.

NO_CHANGELOG=internal
NO_DOC=internal

Methods `get_msgpack` and `dump_msgpack` generally do the same things
except for some details (array header, type of destination buffer), so
let's factor out their core part into a separate function - it doesn't
make much sense right now, but we are going to populate port_c with new
types, and these two functions would become very similar without this
refactoring.

Along the way, let's panic in these methods when we are out of memory -
it's the approach we are taking right now in Tarantool. Corresponding
error injection is dropped as well.

NO_TEST=refactoring
NO_CHANGELOG=refactoring
NO_DOC=refactoring

For easier maintenance, let's panic when there is no memory for allocation
needed to add a value to port - that's the approach we're taking right
now in Tarantool and we don't need to check return code of `port_c_add*()`
functions because they became non-failing. Functions from C API still
return int (which is always zero actually) for backward compatibility.

NO_TEST=no behavior changes
NO_CHANGELOG=no behavior changes
NO_DOC=no behavior changes

Since we have a special mp_ctx type for mp_object, incapsulating
tuple formats and key translation, let's replace iproto_key_translation
with iproto_mp_ctx.

NO_TEST=no behavior changes
NO_CHANGELOG=internal
NO_DOC=internal

We are going to populate port_c with mp_object, which requires mp_ctx.
However, we want to have an ability to dump one port_c instance several
times, so we need to add virtual copy to mp_ctx.

It would be fair to say that copy constructor must accept `src` as a
constant object, but this approach is inconvinient in C, and such method
would be embarrassing to use in our code. So `src` is not constant in
copy constructor.

Method copy of mp_box_ctx is not implemented because it's not trivial,
it probably will require reference counting in tuple_format_map. Anyway,
it is not needed now, so that's not a problem.

Also, mp_ctx_move required dst to be default initialized. However, since
it provides semantics of move constructor, we shouldn't expect dst to be
initialized at all, since constructor is used to construct an
uninitialized object. So let's drop this requirement along the way.

NO_TEST=trivial
NO_CHANGELOG=internal
NO_DOC=internal

Since fiber lua state can be used for internal purposes, we cannot
truncate it, so we need an abitily to create port_lua with lower index
boundary. The patch populates the port with field bottom - dump will
start from this index at Lua stack.

Also, the patch modifies port_lua destructor - all unused values are
popped from lua stack now to prevent Lua stack overflow.

NO_CHANGELOG=internal
NO_DOC=internal

Since we are going to pass arguments to triggers with ports, we need to
move all functions that fire triggers to box because ports are
implemented there.

Making functions that run triggers external (similar to cord_on_yield)
leads to more circular dependencies and problems with linker - I didn't
cope with box and box_error circular dependency, so let's use core
triggers as callbacks to avoid this problem.

NO_TEST=no behavior changes
NO_CHANGELOG=no behavior changes
NO_DOC=no behavior changes

Since we are going to rewrite func_adapter to use ports for arguments
and return values, submodule port must be initialized before `box.cfg{}`
because triggers are allowed to be used before database configuration.
So let's init and free port submodule in `box_init` and `box_free`
functions instead of `box_storage_init` and `box_storage_free` -
`box_storage` is initialized on `box.cfg{}` call.

NO_TEST=no behavior changes
NO_CHANGELOG=no behavior changes
NO_DOC=no behvior changes

We are going keep existing format messages for most of existing error
codes as mentioned in more details in the previous commit. But sometimes
we want error constructor argument to be formatted and to be added to
payload differently. One example is `ER_TUPLE_FOUND` we want tuples to
be formatted as string as before in formatted string but want to add
them as `TUPLE`.

For this purpose let's use "" payload field name. In this case error
constructor argument will be used for formatted string but will not be
added to the error payload.

Follow up #9109

NO_CHANGELOG=internal
NO_DOC=internal

We are going keep existing format messages for most of existing error
codes. This way we will keep all the details in the error messages
until `compat.BOX_ERROR_SERIALIZE_VERBOSE` is switched to 'new' so
details will not disappear from logs/console output. Also we won't need
to change existing Tarantool and integration tests.

Follow up #9109

NO_CHANGELOG=internal
NO_DOC=internal

This commit increases a time to wait of the process termination. It may
take longer than 5 seconds, when tarantool is built with an address
sanitizer. The address sanitizer generates a report at the process
termination and it is not always a fast thing.

NO_DOC=test fix
NO_CHANGELOG=see NO_DOC

Closes #9643

@TarantoolBot document
Title: config: changes in `credentials` section

Now the privileges that were not granted by the configuration, as well
as privileges that were not granted solely by the configuration, are not
revoked on reload. Privileges that have been granted only by the config
module will still be revoked if they are removed from the `credentials`
section on reload.

Currently, if we try to run Tarantool 3.0 with config using old
snapshot, we may get a SCHEMA_NEEDS_UPGRADE error because granting and
revoking privileges are DDL operations. This leads to a situation where
loading Tarantool to perform an upgrade becomes quite problematic. To
avoid the issue, this patch causes 'credentials.lua' to issue a warning
instead of an error in case of the SCHEMA_NEEDS_UPGRADE error during
granting and revoking privileges.

Note that it was still possible to startup and perform the upgrade by
removing the 'credentials' section from the config or without using
config.

This is only a part of the solution, the issue will be fixed in #9849.

Part of #9849
Needed for #9643

NO_DOC=will be added later
NO_CHANGELOG=will be added later

Before this patch, Tarantool, when started with empty configuration
file, used to fail with a non verbose error:
```
$ ./src/tarantool -n instance-001 -c single.yaml
LuajitError: [cluster_config] Unexpected data type for a record: "nil"
fatal error, exiting the event loop
```

Now Tarantool instance can be set up with configuration from remote
configuration sources available in Tarantool Enterprise Edition and
run with empty config file. When there is no configuration provided
for the instance an according error is thrown.

Closes #9845

NO_DOC=bugfix

Currently, the error location (file, line) is always retrieved from
the current stack frame. The new argument `level` allows to change
this. It has the same semantics as the `level` argument of the built-in
Lua function `error`.

Closes #9792

@TarantoolBot document
Title: Document `level` argument of `box.error` and `box.error.new`

Now, when used with a table argument, `box.error` and `box.error.new`
also accept an optional second argument called `level`. It has the same
meaning as the `level` argument of the built-in Lua function `error`:
it specifies how to get the error location. With level 1 (the default),
the error location is where `box.error` / `box.error.new` was called.
Level 2 points the error to where the function that called `box.error` /
`box.error.new` was called; and so on. Passing level 0 avoids addition
of location information to the error.

Example of using `level` with `box.error`:

```lua
local json = require('json')

local function inner(level)
    box.error({message = 'test'}, level)    -- line:4
end

local function outer(level)
    inner(level)                            -- line:8
end

local ok, err
ok, err = pcall(outer)
print(json.encode(err.trace))               -- prints line:4
ok, err = pcall(outer, 1)
print(json.encode(err.trace))               -- prints line:4
ok, err = pcall(outer, 2)
print(json.encode(err.trace))               -- prints line:8
ok, err = pcall(outer, 0)
print(json.encode(err.trace))               -- prints empty table
```

Example of using `level` with `box.error.new`:

```lua
local json = require('json')

local function inner(level)
    local err = box.error.new({message = 'test'}, level)    -- line:4
    return err
end

local function outer(level)
    local err = inner(level)                                -- line:9
    return err
end

local err
err = outer()
print(json.encode(err.trace))               -- prints line:4
err = outer(1)
print(json.encode(err.trace))               -- prints line:4
err = outer(2)
print(json.encode(err.trace))               -- prints line:9
ok, err = pcall(outer, 0)
print(json.encode(err.trace))               -- prints empty table
```

It is also possible to specify `level` when using `box.error` to
re-raise an error created earlier with `box.error.new`, for example:

```lua
local json = require('json')

local err0 = box.error.new{message = 'test'}    -- line:3

local function raise(err, level)
    box.error(err, level)                       -- line:6
end

ok, err = pcall(raise, err0)
print(json.encode(err.trace))               -- prints line:3

ok, err = pcall(raise, err0, 1)
print(json.encode(err.trace))               -- prints line:6

ok, err = pcall(raise, err0, 0)
print(json.encode(err.trace))               -- prints empty table
```

Follow-up #9109

NO_CHANGELOG=internal
NO_DOC=internal

This patch adds a new option '_origin' to the 'box.schema.user.grant()',
'box.schema.role.grant()', 'box.schema.user.revoke()' and
'box.schema.role.revoke()' functions. This is a private option that
allows to differentiate privileges based on their origin. If this option
is not provided an empty string is used by default.

Privileges that were granted using 'grant()' can be revoked using
'revoke()' with the same '_origin'.

The motivation for this change is to avoid actually revoking a privilege
granted by two actors if one actor revokes it.

Needed for #9643

NO_DOC=internal
NO_CHANGELOG=internal

This patch reduces the number of tests in the in2.test.lua test file.
This patch also reduces the number of inserted values. This shouldn't
affect the test since it's not really an original Tarantool test, but it
will reduce the execution time of this test. Currently this test often
fails due to a timeout.

NO_DOC=test
NO_CHANGELOG=test

Prior to this patch, the OP_SetDiag opcode accepted an error code and
error description, which already included the description from the error
code. But this makes #9108 more difficult, so now the accepted error
description does not include the description from the error code.

Note that this change is possible because all accepted error codes use
exactly one argument.

Additionally, this patch removes p4 from the OP_Halt opcode in one
place, since OP_Halt only uses p1 and p2 and does not use p4. Moreover,
the address specified in p4 may not be available by the time OP_Halt is
executed.

Needed for #9108

NO_DOC=internal
NO_TEST=internal
NO_CHANGELOG=internal

* test: set dependencies in BuildTestCLib macro
* Add 'cc' file type for saving bytecode.
* Fix C file generation in jit.bcsave.
* Throw any errors before stack changes in trace stitching.
* Fix recording of __concat metamethod.
* Check frame size limit before returning to a lower frame.
* build: purge sysprof.collapse module
* build: fix tool components handling
* memprof: refactor `heap_chunk` data structure
* memprof: remove unused arguments
* memprof: introduce the `--human-readable` option
* profilers: introduce event reader module
* ci: extend tarantool integration testing

Part of #9595
Part of #5994
Follows up #8700
Needed for #9217

NO_DOC=LuaJIT submodule bump
NO_TEST=LuaJIT submodule bump

After the commit 3daf2399 ("ci: fix step parameters for
reusable runs") it is now possible to create Tarantool
integration workflows for any repository in the Tarantool
organization. Considering this, we don't need to run Tarantool
tests under the old target for the LuaJIT integration in
.test.mk and we can leave only LuaJIT tests in this target for
the sake of exotic Tarantool builds testing.

NO_DOC=CI
NO_TEST=CI
NO_CHANGELOG=CI

Fixes #9680

@TarantoolBot document
Title: config: access configuration of other cluster members

## `config:instances()`

List all instances of the cluster.

Returns a table of the following format.

```lua
{
    [<instance_name>] = {
        instance_name = <...>,
        replicaset_name = <...>,
        group_name = <...>,
    },
    <...>
}
```

If an action should be performed for each instance of the given cluster,
it can be written this way:

```lua
for instance_name in pairs(config:instances()) do
    action(instance_name)
end
```

If replicaset/group names matter, the instances may be filtered this
way:

```lua
-- Perform an action for all instances of the given replicaset.
for instance_name, def in pairs(config:instances()) do
    if def.replicaset_name == box.info.replicaset.name then
        action(instance_name)
    end
end
```

## `config:get(<...>, {instance = '<...>'})`

The new `instance` option of the `config:get()` method allows to
retrieve a configuration value of another instance from the given
cluster.

Example:

```lua
-- Collect all the enabled roles within the cluster.
local enabled_roles = {}
for instance_name in pairs(config:instances()) do
    local roles = config:get('roles', {instance = instance_name})
    for _, role in ipairs(roles) do
        enabled_roles[role] = true
    end
end
```

Note: There is a difference between a missing `instance` option and the
`instance` option that is equal to the given instance name. The former
returns an instance configuration taking into account instance
configuration sources (environment variables). The latter takes into
account only cluster configuration, so the environment variables are
ignored.

Due to a regression introduced in c13b3a31, the worker fiber is started
synchronously, while it should be started asynchronously, in order for the
`wait_connected = false` option of `connect` to work correctly. We already
explicitly wait from Lua for the connection to become active via
`wait_state` when `wait_connected = true`.

Closes #9489

NO_DOC=<bugfix>

Prior to this patch, privileges in the "credentials" section of the
configuration could be requested for non-existent spaces, functions, and
sequences. However, this was not possible for roles, although
essentially the same mechanism is used. The problem is resolved with
this patch.

Part of #9643

@TarantoolBot document
Title: Assign a non-existent credential role

A non-existent credential role can be assigned to a user or role in
`credential` section of the config. The actual assignment will occur
when the assigned credential role is created.

Closes #9809

@TarantoolBot document
Title: New labels in config schema

The new labels are basically maps with string keys and values,
that are merged down to instance level. It means that if a replicaset
and an instance inside it have label with same key and different
value, the actual value is the one instance provided, e.g.

```yaml
groups:
  group-001:
    replicasets:
      replicaset-001:
        labels:
          foo: 'true'
          bar: 'true'
        instances:
          instance-001:
            labels:
              baz: 'true'
              foo: 'false'
```

Results in:

```
./instance-001.iproto> require('config'):get('labels')
---
- baz: 'true'
  foo: 'false'
  bar: 'true'
...
```

If the number of tuple fields is less than `format->min_field_count`,
then some required field is missed, i.e., there is no need to update the
`required_fields` bitmap during msgpack decoding. This optimization is
valid only if tuple format doesn't contain fields accessed by JSON paths.

This patch improves bench_tuple_new by 15-50%, depending on field count.

NO_WRAP
$ taskset 0x2 ~/benchmark/tools/compare.py benchmarks \
  ./tuple.perftest.old ./tuple.perftest.new \
  --benchmark_min_warmup_time=10 \
  --benchmark_repetitions=30 \
  --benchmark_report_aggregates_only=true \
  --benchmark_filter=tuple_new
[...]
Comparing ./tuple.perftest.old to ./tuple.perftest.new
Benchmark                                               Time             CPU      Time Old      Time New       CPU Old       CPU New
------------------------------------------------------------------------------------------------------------------------------------
bench_tuple_new<FORMAT_BASIC>_mean                   -0.1469         -0.1470           126           107           126           107
bench_tuple_new<FORMAT_BASIC>_median                 -0.1428         -0.1429           124           106           124           106
bench_tuple_new<FORMAT_BASIC>_stddev                 +0.0589         +0.0600             4             5             4             5
bench_tuple_new<FORMAT_BASIC>_cv                     +0.2412         +0.2427             0             0             0             0
bench_tuple_new<FORMAT_SPARSE>_mean                  -0.3754         -0.3753          3104          1939          3104          1939
bench_tuple_new<FORMAT_SPARSE>_median                -0.3749         -0.3747          3071          1920          3071          1920
bench_tuple_new<FORMAT_SPARSE>_stddev                -0.3482         -0.3482            85            55            85            55
bench_tuple_new<FORMAT_SPARSE>_cv                    +0.0434         +0.0434             0             0             0             0
NO_WRAP

Needed for tarantool/tarantool-ee#711

NO_DOC=perf improvement

It is possible to skip MP_NIL by mp_decode_nil(), which is faster than
mp_next(). This patch improves bench_tuple_new<FORMAT_SPARSE> by 2.2x.

NO_WRAP
$ taskset 0x2 ~/benchmark/tools/compare.py benchmarks \
  ./tuple.perftest.old ./tuple.perftest.new \
  --benchmark_min_warmup_time=10 \
  --benchmark_repetitions=30 \
  --benchmark_report_aggregates_only=true \
  --benchmark_filter=tuple_new\<FORMAT_SPARSE\>
[...]
Comparing ./tuple.perftest.old to ./tuple.perftest.new
Benchmark                                               Time             CPU      Time Old      Time New       CPU Old       CPU New
------------------------------------------------------------------------------------------------------------------------------------
bench_tuple_new<FORMAT_SPARSE>_mean                  -0.5525         -0.5525          6985          3126          6985          3126
bench_tuple_new<FORMAT_SPARSE>_median                -0.5445         -0.5444          6838          3115          6838          3115
bench_tuple_new<FORMAT_SPARSE>_stddev                -0.8368         -0.8367           541            88           541            88
bench_tuple_new<FORMAT_SPARSE>_cv                    -0.6354         -0.6352             0             0             0             0
NO_WRAP

Needed for tarantool/tarantool-ee#711

NO_DOC=perf improvement
NO_TEST=perf improvement
NO_CHANGELOG=next commit

Implement `class MpData<FORMAT_SPARSE>`, which generates 1000 fields,
10 of them contain unsigned integers, while the remaining are null.

Needed for tarantool/tarantool-ee#711

NO_DOC=perf test
NO_TEST=perf test
NO_CHANGELOG=perf test

Currently `class MpData` generates msgpack data with a predefined format,
let's call it `FORMAT_BASIC`. This patch allows to extend it with other
formats. No functional changes.

Needed for tarantool/tarantool-ee#711

NO_DOC=perf test
NO_TEST=perf test
NO_CHANGELOG=perf test

The test fails when Tarantool with a renewed schema version recovers from
a gh_8937_data/ snapshot, because `luatest/server_instance.lua` executes:
box.schema.user.grant('guest', 'super', nil, nil, {if_not_exists = true})
while insert into system spaces is forbidden until schema is upgraded.
Fix it by adding `box.schema.user.grant()` to the snapshot.

NO_DOC=test
NO_CHANGELOG=test

Fixed a bug when it was possible that the `on_relay_thread_start`
trigger handler would continue to use `data` allocated on the stack
of `box_collect_confirmed_vclock` after returning from it.

Closes #9505

NO_DOC=bugfix
NO_TEST=asan

Closes #9780

@TarantoolBot document
Title: Expansion of `config:info()`

Before this patch, the `meta` field of the `config:info()` function
contained the meta data of the last load, even if it was unsuccessful
reload. This can be inconvenient because if the reload fails, we expect
that the last configuration that was successfully applied is used, but
its metadata was lost.

Now `config:info()` can take one argument — the version of the
information that should be returned by the function. If no arguments are
specified or the argument is the string `v1`, the return value has the
same structure as before. If the argument is the string `v2`, the `meta`
field in the return value has fields - `active` (meta for the last
successfully applied configuration) and `last` (meta of the last loaded
configuration). Currently only these values are accepted.

Bump version of the `codeql-action` action to v3 for fixing an annoying
warning that appears in GitHub WebUI:

    Node.js 16 actions are deprecated.
    Please update the following actions to use Node.js 20

NO_DOC=ci
NO_TEST=ci
NO_CHANGELOG=ci

This patch follows up #9795 where bump of the version of the
`checkout` action was skipped for the etcd_integration.yml and
static_build_pack_test_deploy.yml workflows.

Follows up #9795

NO_DOC=ci
NO_TEST=ci
NO_CHANGELOG=ci

It's impossible to run Node.js 20 actions on CentOS 7 due to issue [1].
So if we want to upgrade to Node.js 20 actions, we have to drop testing
on CentOS 7 in the packaging workflow. Otherwise, the testing job fails
with the following error:

    /__e/node20/bin/node: /lib64/libm.so.6: version `GLIBC_2.27' not found (required by /__e/node20/bin/node)
    /__e/node20/bin/node: /lib64/libstdc++.so.6: version `GLIBCXX_3.4.20' not found (required by /__e/node20/bin/node)
    /__e/node20/bin/node: /lib64/libstdc++.so.6: version `CXXABI_1.3.9' not found (required by /__e/node20/bin/node)
    /__e/node20/bin/node: /lib64/libstdc++.so.6: version `GLIBCXX_3.4.21' not found (required by /__e/node20/bin/node)
    /__e/node20/bin/node: /lib64/libc.so.6: version `GLIBC_2.28' not found (required by /__e/node20/bin/node)
    /__e/node20/bin/node: /lib64/libc.so.6: version `GLIBC_2.25' not found (required by /__e/node20/bin/node)

[1] https://github.com/actions/checkout/issues/1487

NO_DOC=ci
NO_TEST=ci
NO_CHANGELOG=ci

Bump version of the `upload-artifact` and `download-artifact` actions
to v4 for fixing an annoying warning that appears in GitHub WebUI:

    Node.js 16 actions are deprecated.
    Please update the following actions to use Node.js 20

NO_DOC=ci
NO_TEST=ci
NO_CHANGELOG=ci

Let's add an `__autocomplete` method to enable error autocompletion and
make the error object easier to use. It will suggest error object fields
(including own and inherited payload fields) and methods.

Closes #9107

NO_DOC=<not a documentable feature>