Listening, Imagining & Refining: A Heuristic Optimized ASR Correction Framework with LLMs