AI feedback loop: feedback inbox, ai_instructions fix suggestions, and AI evals

[nishantmonu51]

Closes the loop between users rating AI answers and admins improving the project's AI.

Feedback capture and review

• Chat feedback (ratings and explicit review requests) is persisted as ai_feedback rows in a new SQLite migration (0043.sql), including sentiment, categories, comment, AI-predicted attribution and review status.
• New ListAIFeedback, GetAIFeedback and UpdateAIFeedbackStatus runtime APIs, gated on a new ManageAIFeedback permission that is granted to cloud project admins.
• New ListProjectAIFeedback, GetProjectAIFeedback and ResolveProjectAIFeedback local APIs so Rill Developer can review feedback recorded against the project's cloud deployment. Cloud connectivity problems are returned as states (NOT_LOGGED_IN, NOT_DEPLOYED, NO_PERMISSION, ERROR) rather than errors, so the UI can render precise empty states.

Suggested fixes

• GenerateAIFeedbackFix proposes a concrete project change for a feedback item: an ai_instructions rule on the project or a metrics view, or a new measure, along with a draft eval case capturing the exchange.

AI evals

• New eval resource type (parser, reconciler, runner, LLM judge and structural assertions on the metrics view, measures and dimensions the agent queried). Evals never run on a schedule; they are triggered on demand through RefreshTrigger/CreateTrigger, which also supports running a subset of cases and cancelling an in-flight run.
• GenerateAIEvalFix proposes ai_instructions changes that fix the latest run's failing cases without regressing the passing ones.

Frontend

• Feedback inbox with the full conversation transcript and appliable fix suggestions, an eval workspace with a runner and fix panel, and an "add to eval" flow from chat.
• Everything is behind the feedback_inbox and ai_evals feature flags, both default off.

Checklist:

• Covered by tests
• Ran it and it works as intended
• Reviewed the diff before requesting a review
• Checked for unhandled edge cases
• Linked the issues it closes
• Checked if the docs need to be updated. If so, create a separate Linear DOCS issue
• Intend to cherry-pick into the release branch
• I'm proud of this work!