generate image responses to user questions of the EA where appropriate

[jencompgeek]

What is in this PR?

• Adds ability to use gemini-3-pro-image-preview to generate images in response to user questions if preferred and where appropriate
• Uses LLM specified for the conversation to first classify whether visual is appropriate and to actually produce a textual answer to the question. If LLM determines visual response is warranted, gemini uses the generated answer to generate the image (async)
• Adds ability to set user preferences, currently one pref - visualResponses

Changes in the codebase

• Add preferences object to user.model and API methods to get and set preferences
• Add imageGenerator helper and use by all flavors of EA to answer questions when visualResponse pref is set to true
• EA answers questions as usual. If visualResponse pref set, it uses a classification prompt to determine if visual is appropriate. If so, it places the message with the text answer on the image-gen channel (as well as the direct channel so it goes back to the user immediately)
• EA listens on image-gen channel for text response to the user question, then produces a new response message with the image, parentMessage set to the original user question so the UI can flag the relationship.
• Small infra change to allow agents to receive messages from other agents, including themselves
• Add multimodal bodyType to message.model
• Increase websocket maxHttpBuffer to transmit entire base64 image data in a single message
• Update reporting to include the text accompanying the image (blank, but at least gets the message reported) in the direct messages report with a [multimodal] flag to indicate that an image was generated.
• Also includes an unrelated fix to a race condition b/w EA and periodic agents making chat contributions

Documentation and automated testing

Did you:

• document any breaking changes in your commit messages?
• document your changes as comments in the code? Use TSDoc format where appropriate.
• update the README and docs to be clear and easy to use for end users and developers?
• add and/or update automated tests?
• update team documentation of any new or changed environment variables?

Testing this PR

• Ensure GOOGLE_API_KEY and GOOGLE_BASE_URL env variables are set (should be unchanged from previous testing with Gemini models)
• Test in conjunction with FE changes here show images in assistant messages and allow users to set a visual preference nextspace#49
• Create an event assistant plus proactive conversation (any of the EA types should work).
• Launch the assistant page. You should see a dialog on the EA tab allowing you to set a visual preference. Set that preference. Once set, dialog will disappear until you clear cookie and refresh as a different user)
• Ask EA questions that you think should produce a visual response. asking specifically for a timeline, steps in a process, or even just asking it to summarize the event or what you've missed so far should do the trick
• Verify the text response comes back in the usual amount of time with a banner on top showing the question it relates to
• Verify image eventually shows up with the same banner (can try asking a different question in the meantime)
• Reload the page and ensure all messages, including images, are still displayed

Additional information

Opening follow-up ticket to include images somehow in directMessagesReport for evaluation

[bmiller59]

Works! Very cool.