Benchmarking Uncertainty Quantification Methods for Large Language Models with LM-Polygraph