Imagery as Inquiry: Exploring A Multimodal Dataset for Conversational Recommendation