Figure / image ingestion / summarization with model selection

[spencer-crook]

Current app forces /hardcodes Gemini model (which we don't have deployed)

Need to attach currently deployed vision models