Add user feedback reporting to AI messages

Describe the bug
We used to have a user feedback collection button for AI assistant messages, set up here: https://github.com/OpenFn/lightning/pull/2831 . We didn't re-implement it with the React collab-editor AI chat.

Expected behavior
The feedback collection opened a modal with a survey/feedback form, wired to PostHog. However, now that we have Langfuse set up, ideally we'd use that to collect the feedback directly on there. This may need a trace ID, which is currently only handled by Apollo. It might be possible with just the sessionId, which Lightning already handles.

The feedback collection would ideally have thumbs up and thumbs down buttons, and an optional text field. It should identify the specific message the user clicked on, and not the entire session.

We'd need to think through what to do if the user has opted out of metrics, and so there will be no existing session to attach the message(s) to. Additionally, Langfuse is now only tracing core contributor messages, as it isn't self-hosted yet.

Start by reviewing PR 2831 and the current React collab-editor AI chat, then trace how Apollo exposes message or session identifiers and how Langfuse receives traces. Clarify the behavior for metrics opt-outs and untraced messages before implementation. Done means message-specific thumbs up/down feedback with optional text is collected through the chosen service.