LLM4VG: Large Language Models Evaluation for Video Grounding