Replace Skip with AndroidSwiftUI

[colemancda]

Replaces Skip with AndroidSwiftUI. Apple platforms now
use the system SwiftUI directly with no framework in between; Android uses AndroidSwiftUI, selected
per-file with #if canImport(SwiftUI).

Depends on PureSwift/AndroidSwiftUI#4, which adds the SwiftUI surface this app needs
(NavigationStack, TabView, .sheet, .refreshable, lazy stacks, Observation support).

What changed

• Dependencies: skip, skip-model, skip-fuse and skip-fuse-ui are gone, replaced by a
  single AndroidSwiftUI dependency linked only .when(platforms: [.android]). The skipstone
  build plugin, all three Skip/skip.yml files and Skip.env are removed.
• Observation: the model layer already used @Observable/import Observation; the SkipModel
  and SkipFuse imports were purely additive and are dropped.
• Logging: SkipFuse's Logger is replaced by a small AppLogger shim — os.Logger on Apple
  platforms, printing elsewhere.
• App entry: /* SKIP @bridge */ annotations are removed. Darwin/Sources/Main.swift was
  already a real @main SwiftUI App, so nothing was needed to replace the bridge on Apple.
• Skip-only code removed: ComposeView/HeartComposer (Compose interop behind #if SKIP) and
  the unused WelcomeView/"Powered by Skip" template scaffolding, none of which can compile without
  the transpiler. The first tab was labelled "Welcome" with a heart icon while showing the device
  list; it is now "Devices".

Removing Skip also fixed the dependency graph

The package resolves from a clean checkout with no mirrors and no SWIFTPM_ENABLE_MACROS=0, and
the dependency count drops from ~30 to 20. Every blocker documented in Documentation/DependencyState.md
entered through the Skip fork stack:

• the duplicate swift-android-native identity came from skip-android-bridge pointing at a fork
  while PureSwift/Android pointed at upstream — with the bridge gone there is only one claimant;
• the swift-java → swift-syntax 603 vs Bluetooth macros conflict came in the same way;
• skip-fuse-ui no longer constrains skip-ui.

That document is now largely historical. PureSwift/Android#43 and PureSwift/AndroidBluetooth#4 are
still needed — AndroidBluetooth requests AndroidManifest from the wrong package.

Unrelated fix included: plugin deadline

The GATT test suite was failing intermittently. I first blamed machine load; that was wrong — it
failed on an idle machine too. Running each plugin serially showed zero failures, which located
the real cause: the suite loads all 13 bundled plugins per test, each holding a warm interpreter on
its own thread, and running its tests concurrently multiplies that into dozens of contending
interpreters that blow the 50 ms per-call deadline. A healthy plugin was being quarantined for being
slow under contention — which could equally happen on a busy phone during a dense scan.

Two changes: the stress suite is .serialized, and the default per-call deadline is raised from
50 ms to 250 ms. That is still a decisive runaway guard given a real parse takes tens of
microseconds, and warmup is separately bounded at 5 s.

Verification

• Every target builds on Apple platforms with no Skip involvement.
• 44 tests pass, stable across 5 consecutive runs (previously intermittent).
• AndroidSwiftUI's four additions were built together for Android to confirm they compose.

Android is not verified end to end. Before it can be, this needs PureSwift/AndroidSwiftUI#4
merged, the two PureSwift dependency PRs merged, and the Kotlin JNI peers under
Sources/BluetoothExplorer/Skip/ (ScanCallback.kt, BluetoothGattCallback.kt) rehomed into the
Android app project — they are AndroidBluetooth peers, unrelated to Skip, and only lived there
because skipstone collected them. None of the new AndroidSwiftUI code has run on a device;
Documentation/AndroidSwiftUIMigration.md lists its known gaps (no @Bindable, TabView resets
unselected tabs' @State, sheet is a full-screen overlay, .refreshable is inert).