ASR-EC Benchmark: Evaluating Large Language Models on Chinese ASR Error Correction