Not All Experts are Equal: Efficient Expert Pruning and Skipping for Mixture-of-Experts Large Language Models