Mitigating Sensitive Information Leakage in LLMs4Code through Machine Unlearning