After each release, the forge counts /talk turns that resumed their session and those that fell back, and a fallback shows on the vitals #8795

Follows #8776. test/conversation-restart.test.ts (9fa3a56f) proves a resume with the real claude, but it skips without a credential, and a pull's check must never hold one, so the proof never runs there. The forge restarts at every release, which makes production the real test.

Done looks like:

• [x] Each /talk turn records whether it resumed the stored claude_session_id or took the fresh-start fallback from #8754, with the running version. This is one event or column, with no conversation contents.
• [x] The vitals (and MCP flow) show, for the running version, "N conversations resumed after release, M fell back". A fallback after a release is a line on the vitals, not a banner.
• [x] A test with the fake harness: a turn that resumes counts as resumed; a turn fed "No conversation found with session ID" counts as a fallback and appears in the vitals words.
• [x] Nothing gives a pull's check a credential. The real-binary test stays as it is.

Files: src/crew/conversation.ts, src/kernel/vitals.ts, src/flow/vitals.ts, a test beside them (not test/conversation-confirm.test.ts, which is hot) Risks: privacy. Record only the outcome and the version, never the text or the session id in the clear.

Reviewed pull #8804 at its head: ticked Each /talk turn records whether it resumed the stored claude_session_id or took the fresh-start fallback from #8754, with the running version. This is one event or column, with no conversation contents.; The vitals (and MCP flow) show, for the running version, "N conversations resumed after release, M fell back". A fallback after a release is a line on the vitals, not a banner.; A test with the fake harness: a turn that resumes counts as resumed; a turn fed "No conversation found with session ID" counts as a fallback and appears in the vitals words.; Nothing gives a pull's check a credential. The real-binary test stays as it is..