Fix bug: focus can land on a different window of the same app

[dmellok]

The bug

focus on one window of a multi-window app can focus a different window of that app. I hit it constantly with two Safari windows; it also pulled the focused workspace along when the window macOS picked was one AeroSpace had parked off-screen on an inactive workspace.

Conditions, which is why it doesn't reproduce for everyone:

• Displays have separate Spaces off, or a single monitor
• an app with more than one window
• keyboard-driven focus only — a mouse click focuses natively and never enters this path

Cause

MacApp.nativeFocus has a fast path that skips the per-window AX raise and calls nsApp.activate() alone:

if (!NSScreen.screensHaveSeparateSpaces || monitors.count == 1) &&
    (lastNativeFocusedWindowId == windowId || windowsCount == 1)

activate() focuses whichever window macOS currently holds as the app's main one. lastNativeFocusedWindowId records what AeroSpace last observed, and the app's main window drifts from that on its own — a background window taking a link, an app reordering its own windows — so the condition isn't evidence that the requested window is the one activate() will bring.

The general path has a residual race too: the AX raise says which window should be main, and activating the app makes macOS restore whichever window it holds as main.

The fix

Two parts:

1. The fast path now requires windowsCount == 1, where activate() can only mean the requested window. It still covers the case it was added for (apps slow to answer AX requests, which are overwhelmingly single-window).
2. The result is verified rather than assumed: 150ms after asking, check the app's focused window and ask once more if it isn't the one requested. Bounded to a single retry, and skipped when the app is no longer frontmost so that focus moving to another app is left alone.

Measurements

Two Safari windows, focusing a different app between each attempt:

	failures
before	1 in 8
after part 1 only	1 in 8
after both	0 in 24

swift test passes (382 tests).

Happy to split this into two commits, drop the retry and keep only the fast-path fix, or move the discussion to Discussions first if you'd prefer that — just say which.

[dmellok]

Force-pushed a fix for the CI failure. Periphery caught two things:

• raiseAxWindow slipped in from my own fork and doesn't belong in this patch — removed.
• lastNativeFocusedWindowId became assign-only, since the shortcut this changes was its only reader. Removed the property and its assignment in updateFocusCache too, which I should have done in the first place.

All five jobs pass now.