The First-Ever Comprehensive Evaluation Benchmark of Multi-modal LLMs in Video Analysis