ValueBench: Towards Comprehensively Evaluating Value Orientations and Understanding of Large Language Models