FairCode: Evaluating Social Bias of LLMs in Code Generation