iFinder: Structured Zero-Shot Vision-Based LLM Grounding for Dash-Cam Video Reasoning | Read Paper on Bytez