Fix deadlock when handling invalid cached plan errors

[zhanghaitao3]

This commit addresses a deadlock issue occurring when a prepared statement becomes invalid (e.g., after DDL changes).

Key changes:

1. Protocol Safety: Hardened protocol.pyx exception handling to ensure waiter.set_exception() is always called, preventing the executor from hanging indefinitely if exception construction fails.
2. Retry Logic: Updated connection.py to catch generic GaussDBError alongside specific cache errors, ensuring robust detection.
3. Protocol Sync: Added a brief asyncio.sleep(0.05) delay before retrying. This allows the underlying socket to process the ReadyForQuery frame and reset the protocol state properly.
4. Cache Cleanup: Enforced clearing of both local and global statement caches upon error.