PixelRAG & Qwen3-VL: The Next Evolution of Multimodal Retrieval. DeepSeek OCR. Visual Language Model