ToM-LM: Delegating Theory of Mind Reasoning to External Symbolic Executors in Large Language Models