Action Model Learning with Guarantees