T2VSafetyBench: Evaluating the Safety of Text-to-Video Generative Models