Fail the accessibility sweep when a window cannot open

[Blackspirits]

Summary

The new accessibility sweep opens every DI-resolvable tool window and checks its visible inputs, but any exception raised while constructing/showing a window is currently added to the generic skipped list and the test continues. The only coverage floor is opened > 50.

That means a real regression that makes one or many previously covered windows fail during setup can silently remove those windows from the accessibility sweep instead of failing it.

This change:

• keeps intentional structural skips (no one-argument constructor / view model not registered in DI) unchanged;
• records exceptions raised while constructing, showing or initially draining a candidate window as test failures;
• reports the window, exception type and message for diagnosis;
• leaves the existing accessible-name assertions unchanged.

Validation

• based on upstream 45f7430aa7ec8d32f0079563f12ff7c0bfd1dc8e
• upstream PR Accessible names for inputs, lists and grid rows in all tool windows SubtitleEdit/subtitleedit#14825 states that 296 of 299 windows are opened by this sweep
• SkippedWindows is currently empty, so runtime setup failures are not declared expected exclusions
• diff is limited to tests/UI/Logic/AccessibleNamesTests.cs
• full solution restore passed
• full solution build passed
• full solution test suite passed on the first run
• UI suite: 5,150 passed, 9 skipped, 0 failed (5,159 total)
• retry step was not used
• validation run: #34762328881

Because the stricter test remains green, the old exception-to-skip path was not required by the current baseline.

No production behavior is changed.

AI assistance: ChatGPT was used to adversarially review the accessibility test harness for fail-open coverage gaps.

[Blackspirits]

Independent adversarial re-check: the change is test-only and converts runtime window-open exceptions from silent skips into explicit failures while preserving structural skips. Full CI run #34762328881 passed restore, build and the complete solution suite on the first run; UITests reported 5,150 passed, 9 skipped and 0 failed. This also confirms the stricter setup-failure gate is compatible with the current baseline. No blocker identified. Keep draft; no merge or upstream change performed.