"Does the cafe entrance look accessible? Where is the door?" Towards Geospatial AI Agents for Visual Inquiries