*: separate auto_increment ID allocator from _tidb_rowid allocator

[tangenta]

What problem does this PR solve?

Issue Number: close #982

Problem Summary: before this PR, auto_increment ID and _tidb_rowid share the same key-value. The problem is any action that affects auto_increment column also has an impact on _tidb_rowid. For example,

create table t (a bigint auto_increment unique key);
insert into t values (9223372036854775805); -- MaxInt64-2
insert into t values (1);

Query OK, 0 rows affected (0.00 sec)
Query OK, 0 rows affected (0.00 sec)
ERROR 1467 (HY000): Failed to read auto-increment value from storage engine

The first insert statement forces auto_increment ID rebase to 9223372036854775807. Even if the second insert statement does not trigger auto_increment ID allocation, TiDB throws an "Failed to read" error because the _tidb_rowid allocator is exhausted.

What is changed and how it works?

What's Changed:

TiDB behavior changes:

• ALTER TABLE ... AUTO_INCREMENT = ? will not affect _tidb_rowid allocation anymore.
• session variables @@auto_increment_increment and @@auto_increment_offset will not affect _tidb_rowid allocation anymore.
• the function of AUTO_ID_CACHE keeps unchanged(AUTO_ID_CACHE still affects _tidb_rowid allocation).

Related changes

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

Check List

Tests

• Unit test
• Integration test

Side effects

• [ATTENTION] Breaking backward compatibility

Release note

• No release note

[AilinKid]

Rest LGTM

[tangenta]

/run-all-test

[ti-chi-bot]

@tangenta: PR needs rebase.

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.

[tangenta]

/hold because the need to design a complete test

[tangenta]

The new implementation is #28448.