Fix flaky txn_cleanup_range_async_commit_locks test

txn_cleanup_locks_batch_size intentionally leaves locks behind (it
tests that the before-cleanup-locks failpoint prevents actual cleanup).
After scenario.teardown() disables failpoints, the locks remain in
TiKV and can bleed into subsequent tests that assert exact lock counts.

Add an actual cleanup step after teardown so no stale locks persist.

Made-with: Cursor

Author: rushmorem

tests/failpoint_tests.rs

@@ -132,6 +132,16 @@ async fn txn_cleanup_locks_batch_size() -> Result<()> {
     assert_eq!(count_locks(&client).await?, keys.len());
 
     scenario.teardown();
+
+    // Clean up remaining locks to avoid affecting subsequent tests.
+    let safepoint = client.current_timestamp().await?;
+    let options = ResolveLocksOptions {
+        async_commit_only: false,
+        ..Default::default()
+    };
+    client.cleanup_locks(.., &safepoint, options).await?;
+    assert_eq!(count_locks(&client).await?, 0);
+
     Ok(())
 }