feat: trust a resolved conversation as fully handled by support

[lihsaa591]

Summary

New policy: a resolved conversation (first time or Nth time) is trusted as fully handled by support and no longer triggers classification/investigation on its own. The resolved-conversation loop in crisp-classify.mjs now only records "session X seen resolved at time T" into a new state/resolved-seen.json.

Investigation still happens via three paths:

• Manual !tg-autopilot investigate note — unchanged, always overrides, sees the full transcript.
• Reopen after resolve — a conversation that shows up in the active list again after being resolved gets investigated using only the messages after the recorded resolve time. If it's resolved again before a crisp-triage run ever catches it in that in-between window, it's never investigated for that reopen — the resolved-conversation loop just records the newer resolve time and there's nothing left to catch. This is intentional: a second resolve is trusted the same as the first.
• Stale-and-never-resolved auto-escalation — unchanged, for conversations that were never marked resolved at all.

Also updates PHASE2-SETUP.md § 4c and the README's feature table to describe the new policy accurately, and adds state/resolved-seen.json to crisp-triage.yml's committed state files.

Test plan

• node --check passes
• Dry-ran the branching logic (first-resolve records-only, reopen slices to post-resolve messages, manual-note always gets the full transcript regardless, resolved-again-before-caught naturally skips) against mocked data — all four scenarios behaved as intended
• Watch the next few scheduled crisp-triage runs' step summaries for sane counts (resolved conversations fetched vs. actually escalated)