FairMonitor: A Dual-framework for Detecting Stereotypes and Biases in Large Language Models