Towards Unification of Hallucination Detection and Fact Verification for Large Language Models