Fix task_instance_history downgrade failing under the PyMySQL driver

[1fanwang]

The MySQL branch of this downgrade passes two statements to a single op.execute():

op.execute(
    """
    SET @row_number = 0;
    UPDATE task_instance_history
    SET id = (@row_number := @row_number + 1)
    ORDER BY try_id;
    """
)

PyMySQL does not enable CLIENT.MULTI_STATEMENTS, so it rejects the script with a syntax error and the downgrade cannot run. mysqlclient does enable it, which is why the problem is invisible to anyone on that driver, including CI: scripts/ci/docker-compose/backend-mysql.yml uses a bare mysql:// URL, which SQLAlchemy resolves to mysqldb.

Splitting into two op.execute() calls fixes it. op.execute() reuses op.get_bind(), so @row_number is still visible to the UPDATE.

Same class of bug as PR #70235, which fixes the multi-statement SQL in 0017_2_9_2. I found this one while looking at that PR. The two do not overlap.

Testing Done

MySQL 8.4, the migration's own SQL against both drivers, before and after.

Raw logs

Table seeded with try_id out of order (30, 10, 20) so the renumbering is observable:

  CURRENT  OK    mysqlclient  -> [(10, 1), (20, 2), (30, 3)]
  CURRENT  FAIL  PyMySQL      ProgrammingError: (pymysql.err.ProgrammingError) (1064,
                              "You have an error in your SQL syntax; check the manual t...")
  FIXED    OK    mysqlclient  -> [(10, 1), (20, 2), (30, 3)]
  FIXED    OK    PyMySQL      -> [(10, 1), (20, 2), (30, 3)]

The fix produces identical numbering on both drivers, confirming the session variable survives across the two calls and the ORDER BY try_id sequencing is unchanged.