MedFact: Benchmarking the Fact-Checking Capabilities of Large Language Models on Chinese Medical Texts