SF-LLaVA: A Training-Free Video LLM that is Built Upon LLaVA-NeXT and Requires No Additional Fine-Tuning to Work Effectively for Various Video Tasks