FAQ prompts: accept Buddy questions and drop action/mentor requests (chat → Buddy migration)

[daniilperkin]

Parent Story

Part of SprintStartProject/Wiki#319 (chat → Buddy migration). Blocks SprintStartProject/sprintstart-backend#214, which starts feeding Buddy questions into the FAQ. Can be done in parallel with sprintstart-ai#206.

Why

Decision (Wiki#319, 2026-09-25): every question in a project-scoped Buddy conversation feeds the FAQ, in both capability modes; only the AI classifier filters.

Chat only ever received documentation questions. Buddy with capabilities on also gets mentor and action requests:

• "move my card to done"
• "what should I work on next?"
• "remind my PM about the review"
• "summarise my onboarding path"

Both FAQ prompts describe their input as questions "asked to a docs chatbot". Their discard rules only name greetings, smalltalk and thanks, so these requests would become FAQ entries.

Changes

• insights/faq_classification.py _CLASSIFY_SYSTEM, rule 1: also set relevant = false for requests to do something (board actions, reminders, messages to the PM), for questions about the asker's own onboarding state (their tasks, progress, path), and for follow-ups that only make sense inside a conversation ("and the second one?"). Keep it a documentation question when the same docs would answer anyone who asks.
• insights/faq.py _GROUPING_SYSTEM, rule 1: the same, via discard_ids, so a full rebuild agrees with the live classifier.
• Replace "documentation chatbot" / "docs chatbot" with a neutral "the project's assistant" in both prompts, and update FaqClassifyRequest.question's description ("asked in the chat").
• No schema change: the backend keeps sending question text only. It strips quoted selections before sending (backend#214).

Acceptance Criteria

• Action requests, personal-state questions and context-only follow-ups are relevant = false in classify and land in discard_ids in group
• Documentation questions asked in a mentor tone ("how do I get VPN access?") still count
• Tests in tests/insights/test_faq_classification.py and test_faq.py cover each discard category, with ScriptedLLMClient asserting the prompt contains the new rules
• Gates pass: uv run ruff format --check ., uv run ruff check ., uv run pyright src/, uv run pytest