Benchmarking the Text-to-SQL Capability of Large Language Models: A Comprehensive Evaluation