Detecting Misinformation with LLM-Predicted Credibility Signals and Weak Supervision