ddl: add syntax for setting the cache step of auto id explicitly. (#15409)

[sre-bot]

cherry-pick #15409 to release-3.1

In master: the merge order is 
1: auto_id_cache 
2: auto_random_option

Cause auto_random_option has merged first in 3.1, 
so here is a little bit different in code, but no big conflicts in cherry-picking

---

What problem does this PR solve?

That autoid allocator's cache step is quite big now, it will consume a lot of autoid when tidb restarts or crashes. For those scenes which treasure autoid heavily, it's not acceptable.

Parser link: #765

Problem Summary:

What is changed and how it works?

Add auto_increment_cache in create table / alter table statement and it will determine the allocator's cache step.

But there is an exception when a statement like an insert batch(insert into values()()...) which requires allocating consecutive N autoid in one statement, we will make sure that the custom cache step is adequate for it with if step < N, then step = min(2N, maxstep)

Attention please:
tidb handle share the same allocator with auto_increment, so here:
auto_increment_cache will take effect on handle if PKIshandle is false, or even there is no auto_increment column at all.

Related changes

• PR to update pingcap/docs/pingcap/docs-cn:
• PR to update pingcap/tidb-ansible:
• Need to cherry-pick to the release branch

Check List

Tests

• Unit test
• Integration test

Release note:

ddl: add syntax for setting the cache step of auto id explicitly.

[sre-bot]

/run-all-tests

[zimulala]

Please fix the conflicts.
@AilinKid

[AilinKid]

blocked by another parser PR

[AilinKid]

/run-unit-test

[Deardrops]

LGTM

[zimulala]

LGTM

[sre-bot]

/run-all-tests

[djshow832]

LGTM