Modularized Dynamic-Granularity Video LLM for Multi-Event Long Video Understanding