UBENCH: Benchmarking Uncertainty in Large Language Models with Multiple Choice Questions