Dual-Process Scaffold Reasoning for Enhancing LLM Code Debugging