MMCR: Advancing Visual Language Model in Multimodal Multi-Turn Contextual Reasoning