CrEst: Credibility Estimation for Contexts in LLMs via Weak Supervision