Hallucination Detection in Large Language Models with Metamorphic Relations