Describir: Leveraging Static Analysis for Feedback-Driven Security Patching in LLM-Generated Code