[Skeyelab]

Part 2 of #557, stacked on #960. Interface agreed with Eric in the grilling.

• ChannelNaming stays a pure value. mixerOffered(_:) now returns [MixerChange] (.repatched first, then .renamed), and takeStarted(with:) seeds the tracking that was mixerNames/takeRouting in the Recorder. Recorder.offerSources passes each event to the TakeSession and refreshes markers once.
• The saved-name chips move in: saved, copy(_:) (copy from the last Show), persist(to:). Recorder.savedChannelNames is a forward, so App/ is unchanged. The Recorder's public naming methods stay as thin forwards.
• The Recorder still writes the Show's names; type and clear already say whether they changed anything.
• CONTEXT.md: adds Re-patch and Saved name.
• The Recorder loses about 60 lines. Behavior is meant to be identical; the Code Reviewer compared event order, side effects, resets and observation against the removed code, no blockers. I took its points: no separate no-session branch, copy reuses replace, plus tests for a re-patch and a rename in one offer and late names with a change.

Tests (red first: mixerOffered returned nothing, saved did not exist): rename, re-patch, first routing known late, re-patch to an unnamed Source, USB-name fallback, re-patch before rename, late names, chips, copy order, persistence. The existing Recorder and Saved-name tests pass unchanged. Full ./scripts/verify.sh passed on the first version; the review edits were package-only, so after them I re-ran the naming tests and VERIFY_QUICK. Checked by tests and builds only; this touches the Take's rename and re-patch logging, which Eric has not tried on a device.

Draft on purpose: a refactor merges right after a tag. Held until cos says the [internal] tag is out, and after #960 merges.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!