Copy words, Take names, Drive access and Repair each live in one place (#302)

[Skeyelab]

Closes #302

Scope agreed with Eric: "after the Take". Gaps are still found in TakeWriter (on the writer thread, ADR-0002) and tracked by TakeSession during the Take. Nothing in the live recording path changes beyond renaming the access type.

What changed (one commit each)

1. Copy and outcome words live once. DestinationKind.name ("Device"/"Drive"), CopyOutcome.word and TakeMetadata.Repair.Outcome.word. CopySummary, ShowList, ShowReport and CopyCheck build their sentences from these, and every sentence is unchanged. The Show report is per Gap rather than per Copy, so what it shares is the Copy name and the two Repair words. "Not repaired" stays in the report.
2. A Take's name is made in one place. TakeMetadata.name(take:) ("Take 01") replaces 11 String(format: "Take %02d") calls (9 in the Kit, 2 in the App). ReaperExporter.takeName is left alone: it is a Marker label inside ProjectExport, which only renders, not a folder name.
3. One type for the Drive opened for a Take. Destinations.DriveAccess replaces Recording.DestinationAccess. It has a public init(folder:stop:), and release() becomes end(). The App passes store.beginAccess() straight in.
4. DriveAccess can be ended from any thread. An earlier version made end() main-actor only, which broke the App's Show-list load because it ends access inside a detached Task. The tests' release counters are now a Sendable ReleaseCounter.
5. Repair is one module. RepairQueue folds into TakeRepair: an ended Take goes in, Copy outcomes come out. The work that fills Gaps (fill, length, bytesToExtend) is now internal to it. StemRepair in BroadcastWave is untouched.

Kept on purpose

The run: test seam on TakeRepair stays. The real Gap fill and the tests' stand-in are two adapters. The stand-in's gate is the only test that pins "the latest Take's outcome wins" deterministically: RepairTests.laterTakeOutcomeWins passes whichever order the Repairs finish in.

How it was checked

• New tests:
  • every CopySummary word, pinned on the old code before the refactor
  • TakeMetadata.name(take:), written first, failed before the function existed
• Existing pins kept green: ShowListTests and CopyCheckTests sentences, and TakeFileFormatTests, which covers the report's per-Gap words.
• "Repairs as today": RepairTests "After a Drive fails and comes back, Repair fills the Gap and both Copies end byte-identical", plus the rest of RepairTests, unchanged and green.
• ./scripts/verify.sh passes after rebasing onto main with Record at the input's sample rate, 96 kHz included (#195) #308, Release notes for v0.1.19 #309 and Sheets follow a new Appearance at once #310: 426 package tests, macOS and iOS Simulator builds.
• Not run on a device. Repair still writes files. The TestFlight check from The Recorder checks its Destinations itself and says why a Take ended (#299) #307 (pull and replug the Drive mid-Take) also covers this.

https://claude.ai/code/session_013qdugNZbXwKKz5gPzzi2CL