MME-CoF-Pro: Evaluating Reasoning Coherence in Video Generative Models with Text and Visual Hints — arXiv2