Extract MainWindowViewModel from MainWindow

[danielchalmers]

What

Splits the 996-line MainWindow god-object into a dedicated MainWindowViewModel plus a thin view shell, then adds unit tests for the extracted logic.

Before

MainWindow.xaml.cs (996 lines) was simultaneously the Window, the view-model ([INotifyPropertyChanged] + DataContext = this), the playback orchestrator, the update checker, the FFmpeg prompt host, the clip loader, the keyboard handler, and the shell-action host. None of that logic was unit-testable, because you can't instantiate a WPF Window in a test.

After

• MainWindowViewModel.cs (new) — all observable state, computed properties, commands, clip loading, playback orchestration, update checks, FFmpeg prompts, the error-overlay state machine, and keyboard handling.
• MainWindow.xaml.cs (996 → 227 lines) — only genuine view concerns: window lifecycle/shutdown, the Flyleaf host-layout engine (reparents named controls), seek-slider input plumbing, search-box focus, and hyperlink navigation.
• MainWindow.xaml — only the design-time d:DataContext type changed.

Seams used to keep them decoupled

• The view injects a Func<VideoPlayerController> factory closing over the named FlyleafHost controls, so the VM owns the controller without referencing any control.
• The VM raises SearchBoxFocusRequested; the view handles focus.
• The view reacts to SelectedCameraView changes to drive host layout.

Tests

Adds MainWindowViewModelTests (27 new cases) covering logic that previously had zero coverage: camera-view selection, about toggle, the overlay/visibility state machine, play/stop enablement, loading/render text, the update badge, the search-focus keyboard shortcut, and error dismissal.

Full suite: 81 passing (was 54).

Behavior

No intended runtime change. Verified locally: builds clean, all tests pass, and a startup smoke-run produces the identical initialization sequence (init → FFmpeg check → prompt). Manually confirmed working by the maintainer.

Notes / follow-ups

• FilteredClips filtering, seek math, and the clip→auto-play path need a small DI seam or a pumped dispatcher to test cleanly — deferred.
• This unblocks further slicing (ClipBrowser / Playback / Updates / About sub-models) and is step 1 of the agreed reorg plan.