Part of #6773

NO_DOC=Will be added later.
NO_CHANGELOG=Will be added later.

This patch introduces basic INTERVAL support in SQL. After this patch,
it will be allowed to select INTERVAL values from spaces, insert them
into spaces, and use them in functions. CAST() from INTERVAL to STRING
and ANY is also supported.

Part of #6773

NO_DOC=Will be added later.
NO_CHANGELOG=Will be added later.

This patch fixes the error description for comparison in case the left
operand is comparable but the right operand is not.

Part of #6668
Needed for #6668

NO_DOC=Will be added in another commit.
NO_CHANGELOG=Will be added in another commit.

After this patch, using MAP and ARRAY in GROUP BY will throw an error
instead of an assertion.

Part of #6668
Needed for #6773

NO_DOC=Will be added in another commit.
NO_CHANGELOG=Will be added in another commit.

After this patch, when DISTINCT is used with MAP or ARRAY, an error will
be thrown instead of a segmentation fault.

Part of #6668
Needed for #6773

NO_DOC=Will be added in another commit.
NO_CHANGELOG=Will be added in another commit.

LuaJIT submodule is bumped to introduce the following changes:
* memprof: enrich symtab with newly loaded symbols
* memprof: extend symtab with C symbols

Within this changeset the new Lua module providing minimalistic AVL tree
implementation required for processing C symbols is introduced:
* utils/avl.lua: minimalistic AVL tree implementation

Closes #5813

NO_DOC=LuaJIT submodule bump
NO_TEST=LuaJIT submodule bump

Signed-off-by: Igor Munkin <imun@tarantool.org>

New field type INTERVAL is now supported by BOX.

Part of #6773

NO_DOC=INTERVAL has already been introduced earlier.

Part of #6773

NO_DOC=INTERVAL has already been introduced earlier.
NO_CHANGELOG=It will be added in another commit.

Part of #6773

NO_DOC=INTERVAL has already been introduced earlier.
NO_CHANGELOG=It will be added in another commit.

Part of #6773

NO_CHANGELOG=It will be added in another commit.

@TarantoolBot document
Title: New MsgPack extension type MP_INTERVAL

MsgPack extension type MP_INTERVAL was created to serialize values of
struct interval.

Description of serialized value:

------------------------------------------------------------
|MP_EXT|Size of packed interval|MP_INTERVAL|packed interval|
------------------------------------------------------------

Packed interval - packed number of non-zero fields and packed non-null
fields. Each packed field consists of two packed integer values - the
field ID and its value. The number of non-null fields can be zero, in
which case `packed interval` will be just packed integer `0`.

List of IDs:
0 - `year`
1 - `month`
2 - `week`
3 - `day`
4 - `hour`
5 - `minute`
6 - `second`
7 - `nanosecond`
8 - `adjust`

Example of packed interval value:

|C7|09|06|03|00|01|01|CCC8|03|D0B3|

Where
C7 - MP_EXT;
09 - size of packed interval value, 9 bytes;
06 - MP_INTERVAL;
03 - number of non-zero fields;
00 - ID of field `year`;
01 - packed value `1`;
01 - ID of field `month`;
CCC8 - packed value `200`;
03 - ID of field `day`;
D0B3 - packed value `-77`.

From there we can see, that the packed interval is actually
`1 year, 200 months and -77 days`.

This method is necessary to compare two INTERVAL values in Luatest.

Part of #6773

NO_DOC=Internal function.
NO_CHANGELOG=INTERVAL has already been introduced earlier.

Half of allocation functions in space_def_dup and space_def_new already
use xmalloc. Use xmalloc for the rest of allocations.

NO_DOC=refactoring
NO_TEST=refactoring
NO_CHANGELOG=refactoring

To implement space upgrade, we need a function that duplicates field_def
array, because we will store a new space format in space options.

To factor out field_def_array_dup from space_def_dup, let's allocate it
with a separate malloc rather than after space_def struct.

NO_DOC=refactoring
NO_TEST=refactoring
NO_CHANGELOG=refactoring

NO_DOC=internal
NO_CHANGELOG=internal

NO_DOC=refactoring
NO_TEST=refactoring
NO_CHANGELOG=refactoring

The function isn't used since commit 3dcc3fd6 ("Move
default_value_expr from field_def to tuple_field").

NO_DOC=refactoring
NO_TEST=refactoring
NO_CHANGELOG=refactoring

We have found that 301-basic.t from Lua harness tests became
flaky on macosx runners.
Please see https://github.com/tarantool/tarantool/issues/7058

 for
more details. Use simple workaround `collectgarbage('collect')`
for temporary workaround of this problem.

Co-authored-by: Igor Munkin <imun@tarantool.org>

NO_DOC=infra
NO_CHANGELOG=infra
NO_TEST=infra

Part of #6923

NO_TEST=docrequest

@TarantoolBot document
Title: Document interval arithmetic

Interval arithmetic
-------------------

If we need to shift date values by the given period of time we call either
modifier methods (i.e. `:add` or `:sub`) or apply interval arithmetic
using overloaded `+` (__add) or `-` (__sub) methods.

`:add`/`:sub` modify current object, but `__add`/`__sub` create copy of
object for operation result.

In the interval operation we sequentially calculate each interval
subcomponents
heading from largest (year) to smallest (nanosecond):
* `year` - year(s);
* `month` - month(s);
* `week` - week(s);
* `day` - day(s);
* `hour` - hour(s);
* `min` - minute(s);
* `sec` - second(s);
* `nsec` - nanosecond(s).

If results of operation exceed allowed range for any of component then
exception to be raised.

Modifier methods `:add`/`:sub` return `self` object thus it''s possible to
chain their calls together, i.e.

```
-- add 9000 years, 82 months, 5 weeks, 201 days, 183 hours, 292 minutes
-- and 191.001239234 seconds to given date
dt:add{
	year  = 9000,
	month = 82,
	week  = 5,
	day   = 201,

	sec   = 191,
	min   = 292,
	hour  = 183,

	nsec  = 1239234,
}

dt:sub{
	year  = 9000,
	month = 82,
	week  = 5,
	day   = 201,

	sec   = 191,
	min   = 292,
	hour  = 183,

	nsec  = 1239234,
}

-- chaining
dt:add{year = 2}:add{month = 2}:sub{day = 2}

-- create a copy of current date, while moving it to the next day
local dt2 = datetime.new(dt:totable()):add{day = 1}
```

Arithmetic operations
=====================

Date and interval objects may participate in arithmetic operations:
* Sum of 2 intervals is interval object, which fields would be sum of each
  particular components of operands;
* Subtraction is similar to summation above - result is interval object
  where each subcomponent would be result of subtraction of particular
  fields in original operands;
* If you add date and interval then result is a date. Addition to be
  performed in a determined order from largest component (i.e. year) to
  smallest (nanosecond);
* Subtraction of dates to produce an interval object. Difference of 2 time
  moments is performed not as difference of epoch seconds, but as
  difference of all subcomponents (i.e. years, months, days, hours,
  minutes and seconds);
* Untyped table object may be used in each context where typed date or
  interval object used if left operand is typed object with overloaded
  operation of '+' or '-'.

Matrix of *addition* operands eligibility and their result type:

|                     | datetime | interval | table    |
|---------------------|----------|----------|----------|
| **datetime**        |          | datetime | datetime |
| **interval**        | datetime | interval | interval |
| **table**           |          |          |          |

Matrix of *subtraction* operands eligibility and their result type:

|                     | datetime | interval | table    |
|---------------------|----------|----------|----------|
| **datetime**        | interval | datetime | datetime |
| **interval**        |          | interval | interval |
| **table**           |          |          |          |

Date adjustions and leap years
==============================

It''s always tricky to operate with days if we move date between months of
different lengths. The default mode is - the day number in month of
original date should become the same day but not exceed the length of
resultant month.

* 28 February of non-leap year if added +1 year should become 28 february
  of a leap year;
* Addition of year to the 29 February of a leap year would result with 28
  February of a non-leap year;

The general rule is as following: addition of months to the date should
produce (if possible) the same day number, if this number is not exceeding
number of days in resultant month. Otherwise it produces the last day in
month.

* 31 january + 1 month = 28 or 29 february;
* 30 january + 1 month = 28 or 29 february;
* 29 february + 1 month = 29 march;
* 31 march + 1 month = 30 april.

Optionally you may prefer to use "accountant" logics of operations when
month operations (addition or subtraction) should leave last day in month
as last day, i.e.:

* 31 january + 1 month = 28 or 29 february (depending on leap year);
* 29 february + 1 month = 31 march;
* 31 march + 1 month = 30 april;
* 30 april + 1 month = 31 may.

- 28 february 2001 + 1 month = 28 march 2001;
- february 2004 + 1 month = 28 марта 2004;

This adjustment option named `adjust` may be passed as additional argument
to `:add`/`:sub` methods or additional argument to interval constructor
call.

There are 3 supported `adjust` values:

* `none` last day mode is not maintained, if there is overflow then value
  is truncated to the maximum possible value in month (corresponding to
  the c-dt `DT_LIMIT`). This is default mode if not provided;

```
    -- 29.02.* -> 29.03.*
    dt:add( {month = 1, adjust = "none" } )
```

* `last` last day mode ("accountant" mode) maintained. i.e. last day in
  original month should stay last day in resultant month. This corresponds
   to `c-dt` `DT_SNAP` mode;

```
-- 28.02.2001 -> 31.03.2001 -- last day in Feb 2001
-- 28.02.2004 -> 28.03.2004 -- not a last day in Feb => no abjustments
-- 29.02.2004 -> 31.03.2001 -- last day in Feb 2004
dt:add( {month = 1,  adjust = "last" } )
```

* `excess` overflow to the next month performed. According to `c-dt`
  `DT_EXCESS` rules.

Stringization
=============

Intervals show each their subcomponents as negative or positive value.
Seconds and nanoseconds normalized to the human-interpretable form.

* '+1 seconds';
* '+12 hours, 10 minutes, 30 seconds';
* '-20 months, -10 weeks, -8 hours, -10 minutes, -30 seconds';
* '-5000000 years, -20 months, -10 weeks, -10 minutes, -30 seconds';

We may overflow supported years range if we proceed
(multiple times) interval addition or subtraction operations.
At the end of add/sub we check all interval subcomponents
on validity of their value. Result of this check is returned
in the encoded value to the Lua side, where less than 0 means
underflow in some component, and greater than 0 overflow in
subcomponent.

Part of #6923

NO_DOC=fix
NO_CHANGELOG=fix

Split single sec field in the datetime_interval into multiple subfields,
i.e. week, day, hour, min, sec. Make sure that whatever we enter in the
constructor we get similarly visualized as textual representation

Lua implementation of datetime_totable has been converted to C, which
allows to use it for datetime values decomposing in datetimes
subtraction. It uses per component subtraction for calculating of a
resultant interval.

Part of #6923

NO_DOC=refactoring
NO_CHANGELOG=refactoring

Created C implementation for the rest of datetime/interval operations
and call them via FFI. We need it for later Olson integration.

Part of #6923

NO_DOC=refactoring
NO_CHANGELOG=refactoring
NO_TEST=refactoring, tests left intact

Convert Lua implementation of datetime_increment_by to C from their
Lua variant.

Part of #6923

NO_DOC=refactoring
NO_CHANGELOG=refactoring
NO_TEST=refactoring, tests left intact

Part of #6923

NO_DOC=refactoring
NO_CHANGELOG=refactoring
NO_TEST=refactoring, tests left intact

Part of #6923

NO_DOC=internal
NO_CHANGELOG=internal
NO_TEST=refactoring, tests left intact

Simplify datetime interval data handling in Lua module:
* everytime we used to pass year, month, day, hour and second we simply
  pass aggregate containing those values.

Part of #6923

NO_DOC=refactoring
NO_CHANGELOG=refactoring
NO_TEST=refactoring, tests left intact

This patch fixes an error that occurred when a UUID index was created
on a SCALAR field or a SCALAR index was created on a UUID field.

Follow-up #6042

NO_DOC=Follow up.
NO_CHANGELOG=Follow up.

So that we can reuse it for online space upgrade.

NO_DOC=refactoring
NO_TEST=refactoring
NO_CHANGELOG=refactoring

This is needed to factor out the function that decodes space format so
that it can be used for online space upgrade. It's not a big deal,
because we already use ER_WRONG_SPACE_OPTIONS if space options specified
on space creation/alter are wrong so using ER_WRONG_SPACE_FORMAT would
only be consistent.

NO_DOC=minor change in error message
NO_CHANGELOG=minor change in error message

Currently, we use ER_WRONG_INDEX_OPTIONS instead, which is confusing.
Also, let's simplify the error message: remove the reference to the
deprecated index parts format.

NO_DOC=minor change in error message
NO_CHANGELOG=minor change in error message

opts_decode takes errcode and field_no, which is passed to diag_set,
so all code paths that use opts_decode must set a field_no in an error
message. This is stupid, because often there's no reasonable field_no to
report. As a result, for ER_WRONG_INDEX_OPTIONS, ER_WRONG_SPACE_OPTIONS,
and ER_WRONG_COLLATION we use either 0 or field_no of the field in a
system space, which means nothing to users. When there's an error in
index parts we pass the index part no for field_no ("(field X)" where X
is an index part number), which is confusing. The only error for which
field_no makes sense is ER_WRONG_SPACE_FORMAT, but it doesn't justify
passing around field_no for other errors and complicating common code
paths.  Let's drop the field_no everywhere for simplicity - the user can
pin-point the erroneous field without it by inspecting fields/parts.

NO_DOC=minor change in error message
NO_CHANGELOG=minor change in error message

Options created with space_opts_decode are never explictily freed,
because they are supposed to use only region memory. For some reason,
we duplicate space_opts.sql in space_opts_decodes, which results in
a memory leak. Remove this code.

NO_DOC=bug fix
NO_TEST=memory leak
NO_CHANGELOG=minor bug

Due to full-ci a few problems were found.

NO_DOC=Fixes for fill-ci
NO_TEST=Fixes for fill-ci
NO_CHANGELOG=Fixes for full-ci

Part of #6773

@TarantoolBot document
Title: SQL built-in function DATE_PART()

The DATE_PART() function returns the requested information from a
DATETIME value. It takes two arguments, the first one tells us what
information is being requested, the second is a DATETIME value. Here is
a list of supported values of the first argument and what information is
returned:
'millennium' - milennium
'century' - century
'decade' - decade
'year' - year
'quarter' - quarter of year
'month' - month of year
'week' - week of year
'day' - day of month
'dow' - day of week
'doy' - day of year
'hour' - hour of day
'minute' - minute of hour
'second' - second of min
'millisecond' - millisecond of second
'microsecond' - microsecond of second
'nanosecond' - nanosecond of second
'epoch' - epoch
'timezone_offset' - timezone offset

Part of #6773

NO_CHANGELOG=It will be added in another commit.

@TarantoolBot document
Title: SQL built-in function NOW()

The NOW() function returns the current date and time as a DATETIME
value. If this function is called more than once in a query, it returns
the same result until the query completes, unless a yield has occurred.
On yield the value returned by NOW() changes.

This patch enables creation of a space with a foreign key pointing
to the same space. Such spaces are marked with a `selfpin` flag in
the `space_cache_pin_list`. If a space is pinned only by selfpins,
it is allowed to be deleted from the space cache.

When an old space is replaced by the new one, it's self-pinned for the
second time, then all old pins are moved to the new space's pin list
(see `space_cache_repin_pinned`), then the first self-pin is removed
by `space_cleanup_constraints` of the old space.

Closes #6961

NO_DOC=bugfix

Part of #6961

NO_DOC=bugfix
NO_CHANGELOG=see later commits
NO_TEST=see later commits

It's always false.

I don't understand why it was added in the first place. Initially, it
was introduced by commit b9c9792d ("sql: move default col values to
Tarantool's core"). The flag was set to true only in Tarantool core for
some reason. Then there was commit 7c6c572c ("sql: Fix DEFAULTs AST
memory leak on alter") that claimed there was a memory leak because the
flag was set it to true and switched it to false in all places except
one: when a space_def was destroyed it was still set to true. I dropped
usage of space_expr_dup and switched the flag to false in the previous
commit, and everything seems to work just fine now.

NO_DOC=refactoring
NO_TEST=refactoring
NO_CHANGELOG=refactoring

field_def, like any other _def struct, is supposed to be plain and
simple - it should map data stored in the _space system spaces as is.
Complex data structures produced from it, like field dictionary or
compiled SQL expression should be stored in tuple_field. This simplifies
reasoning about the lifetime of this structure:
 - If it's allocated on the region, as it is the case, when the
   structure is created from MsgPack data, it will live until the
   region is truncated. It doesn't need to be destroyed in this case.
 - If it's allocated on malloc (by a dup function), it should be
   explicitly deleted with a delete function.

Why we are doing this now: to implement space upgrade, we'll need to add
a new member to space_opts struct. The new member will be a struct that
may store an array of field_def. The problem is space_opts created on
the region from MsgPack data is never deleted (see alter.cc), which
means that the expression stored in field_def could leak.

Note: in order to avoid tuple -> sql dependency, we have to call
sql_expr_compile and sql_expr_delete from tuple_format.c via a callback
set by sql_init.

NO_DOC=refactoring
NO_TEST=refactoring
NO_CHANGELOG=refactoring

When processing promote/demote row from wal in relay, process tx
endpoint until last sent raft term is greater or equal to term in
promote/demote request.

Part of #6754

@TarantoolBot document
Title: box.info.synchro.queue.term added

Added `term` field to `box.info.synchro.queue`. It contains term of the
last PROMOTE. It is usually equal to `box.info.election.term` but may be
less than election term when new round of elections started, but no one
promoted yet.

As stated in its comment, this routine was needed back in the time when
raft followers ignored data coming from a non-leader. This is not the
case anymore. Now the followers accept the data from any source, so
there's no need to try and re-send some data which accumulated while the
node hasn't yet broadcast its leadership.

Part of #6754

NO_DOC=dead code removal
NO_CHANGELOG=dead code removal
NO_TEST=dead code removal