executor: add a variable to control whether we can write _tidb_rowid

[zimulala]

What problem does this PR solve?

There are problems handling the following case:

create table tt(id binary(10), c int, primary key(id));
insert tt values (1, 10);
insert into tt (id, c, _tidb_rowid) values(30000,10,1);
tidb> select * from tt use index(`primary`);
+------------+------+
| id         | c    |
+------------+------+
| 30000      |   10 |
| 30000      |   10 |
+------------+------+
2 rows in set (0.00 sec)
tidb> select * from tt;
+------------+------+
| id         | c    |
+------------+------+
| 30000      |   10 |
+------------+------+
1 row in set (0.00 sec)

What is changed and how it works?

At the moment, if we have to deal with _tidb_rowid, it will be more complicated. So it is best not to support inserting, updating, and replacing _tidb_rowid statements now. But we have some tools that need to write _tidb_rowid, these tools will guarantee that they will not appear as above.
Currently, we want users not to open AllowWriteRowID.

Check List

Tests

• Unit test

Related changes

• Need to cherry-pick to the release branch

[zimulala]

/run-all-tests

[morgo]

I believe this might break https://github.com/pingcap/mydumper 's restore.

PTAL @kennytm

[kennytm]

Could we have an internal flag which allows INSERT to accept a _tidb_rowid column? Lightning requires _tidb_rowid to ensure idempotence when importing the same part of the table twice after resuming from a lagged checkpoint. Maybe Loader needs this too.

[zimulala]

/run-all-tests tidb-test=release-2.0 tidb-private-test=release-2.0 tikv=release-2.0

[WangXiangUSTC]

in tidb-binlog, we use _tidb_rowid to guarantee the idempotence when table's PKIsHandle is false. same with lightning and loader.

[winkyao]

LGTM

[zimulala]

PTAL @jackysp

[zz-jason]

LGTM