Add a switch to the per-display connection row

[Blazetes]

What & why

Replace the one-way Disconnect Display action with a native Display Connection switch at the same position in the display detail. Held-disconnected displays use an off-state switch to reconnect. The display header continues to expand its settings.

The controls reuse physical-disconnect and input-aware reconnect services, show failures and pending progress, and protect the last viewable physical screen. They remain hidden on Intel and Crisp virtual displays. A shared transaction gate stays held until the blocking configuration really finishes, even after the UI timeout; queued requests pin their UUID, recheck safety after DDC drains, and retain late-completion recovery state.

The physical-configuration safeguards are shared with the related batch-controls PR, #209. Both feature branches are independently buildable; after either merges, the other can be rebased to remove the shared backend portions.

How tested

• Exact source 8570f4cc4ab9d19984c447dabbd379ef0b4212f5 passed local make check: full strict SwiftLint, unit tests, x86_64 typecheck and all 274 extracted localization keys.
• Simplified/Traditional Chinese completeness and the full local ./scripts/release.sh v0.0.0-ci universal DMG build passed with Xcode 27.0 (27A266a) / SDK 27.0.
• The application and CLI contain arm64/x86_64 slices; deep signature and DMG integrity checks passed.
• Regression tests cover pending transaction exclusivity, timeout safety, actual-completion release and queued identity capture. Three-language native fixtures compile the production views against inert display services.
• Live display switching, clamshell/sleep recovery and installed-app interaction: not run.

Checklist

• Builds locally (./scripts/release.sh v0.0.0-ci, exact source above)
• Added/updated Crisp/Resources/Localizable.xcstrings
• Screenshot or short clip for UI change

Native SwiftUI fixtures, not live-monitor screenshots:

English	Simplified Chinese

[didriksg]

I'm closing this one, but the switch is coming in another shape, in #218. It sits on the display row itself where the chevron is, behind a setting that's off by default, and it uses the existing disconnect and reconnect unchanged. A switch inside the expanded card still leaves the control one click deep, and that was the complaint in #103.

I'm leaving out the configuration gate and the late-completion handling. They change a large part of the disconnect path, which I'm keeping frozen until I have field data, and they weren't run on hardware. Reading them is how I found a real race, though: two disconnects close together could both pass the last-screen guard. #218 closes it by running connection changes one at a time.