Investigating Video Reasoning Capability of Large Language Models with Tropes in Movies