simulator: checkpoint after an open savepoint fails the run with TableLocked

The simulator (Faultless) job failed on #9136.

• Run: https://github.com/tursodatabase/turso/actions/runs/35269350663/job/105364728207
• Seed: 2203329238356856833
• Error: failed with error: 'TableLocked'

What the log shows

1. Connection 5 runs SAVEPOINT sim_sp_419;. This opens a transaction.
2. The next statement, DELETE FROM funny_crudo_231 WHERE TRUE;, fails with Busy.
3. is_recoverable_tx_error (testing/simulator/runner/execution.rs:27) accepts Busy, so the run continues. The savepoint stays open.
4. Later, connection 5 runs PRAGMA aux0.wal_checkpoint(FULL);.
5. op_checkpoint (core/vdbe/execute.rs:692) returns TableLocked, because the connection is still in a transaction. SQLite does the same.
6. TableLocked is not in the recoverable list, so the run fails.

The engine behavior is correct. The simulator must close the savepoint after a Busy, or accept TableLocked as a legal result.

The PR changes only hash join selection. The failure is in other code.