Enhancing LLM Reasoning with Multi-Path Collaborative Reactive and Reflection agents