refactor: extract app-target orchestration logic into WhiskyKit for testability

[frankea]

About 1,170 lines of real orchestration logic currently have no tests because they live in the app target, which swift test can't reach:

File	Lines	Untested logic
Whisky/Utils/SteamClientOrchestrator.swift	~300	per-app launch phases, client-startup single-flight, shared process snapshot, launcher-fix application
Whisky/Utils/LauncherDetection.swift	~330	launcher fix application and profiles (detection itself moved to WhiskyKit in #160)
Whisky/Extensions/Bottle+Extensions.swift	~540	bottle move/export/duplicate state machines

This is not hypothetical debt. The bottle-move corruption (#154), the in-flight guard loss (#155), and the launcher-fix bypass fixed in #166 all lived in these files, and every one shipped without a test because none was possible. Review keeps catching bugs here for a structural reason: this is exactly the code external contributors cannot compile locally (the kit builds with Command Line Tools alone; the app target needs full Xcode), so the code with the least pre-PR verification is also the code with zero CI test coverage.

Plan, in extraction order (each step is independently shippable, no behavior change intended):

1. Bottle operations into WhiskyKit: the move/export/duplicate logic behind a small seam for the BottleVM interactions (the pin/blocklist rewrite-and-rollback, the in-flight lifecycle). The fix: dont corrupt bottle state on failed move #154/fix: keep in-flight bottles alive across registry reloads #155 fixes get regression tests at last.
2. Launcher fix application into WhiskyKit next to LauncherType.detect: the profile table and the apply/refine rules (launcherMode gating, already-configured short-circuit). The fix: steam library follow-ups from review #166 gating semantics get pinned.
3. Orchestrator core into WhiskyKit: the phase dictionary, single-flight client startup, snapshot TTL logic, parameterized over a process-runner protocol so tests can drive it without wine. The SwiftUI view keeps only observation and presentation.

Acceptance: the three app files become thin adapters, the extracted logic has kit tests covering the bugs we've already fixed there, and codecov's whiskykit flag picks the new files up automatically.

[frankea]

Progress: step 1 (bottle operations) merged in #175, step 2 (launcher fix application) merged in #176. Together they moved roughly 700 lines of previously untestable state machine and settings mutation logic into WhiskyKit, now covered by 22 dedicated tests on real on-disk bottles, including regressions for the #154 rollback and the #166 gating semantics. Step 3 (the orchestrator core) stays parked until the steam surface settles; this issue tracks it.

[frankea]

Step 3 landed in #241, which completes the plan here.

Final tally across the three steps: bottle operations (#175), launcher fix application (#176) and now the Steam client orchestrator core (#241) live in WhiskyKit, each behind a small seam, each with kit tests that reach the bugs that motivated this (#154, #155, #166, and the client-startup and process-snapshot races the orchestrator exists to prevent). The orchestrator file itself measures 98.9% line coverage under swift test; what remains in the app target is a 30-line driver subclass for download monitoring, the SwiftUI observation in LibraryModel, and the four calls that need a running Wine.

Closing.