Probing the "Conscious" Mind: Selective Access and Global Workspace in Vision-Language Models — ThinkLLM