test(045): tear-off E2E suite reliable under WinAppDriver (#419) (#426)

* test(045): tear-off E2E suite reliable under WinAppDriver (#419)

The 3 [Ignore]'d tests in DockingTearOffE2ETests.cs (E02/E03/E04) tried
to drive cross-window float→host drags that turn out to be blocked at
the WinAppDriver pipeline level: when the source floating window is
hidden mid-drag (single-tab BeginFloatingTearOff → AppWindow.Hide()),
the session-bound input pipeline freezes and subsequent moves never
reach the host's overlay PointerEntered handlers — verified via
detailed Tracker.OnTick / Overlay.PointerEntered tracing. Pure-Win32
SendInput has its own coherence problem (test-process and host-process
GetAsyncKeyState(VK_LBUTTON) disagree mid-drag, likely UIPI hook
interaction).

Replace those tests with a reliable 4-test suite that exercises the
WinAppDriver-tractable scenarios end-to-end with real mouse input.
12/12 green across 3 consecutive runs:

- E01: single-tab tear-off → floating window (drop-outside).
- E02: multiple sequential tear-offs (A then B → two floating windows,
  asserts OnContentFloating/Floated fire exactly twice).
- E03: pane content state survives the tear-off layout mutation.
- E04: tear-off pipeline reliable across 3 repeated invocations.

Cross-window dock-back / split / Esc-cancel scenarios remain covered
by NativeDockingTearOffFixture's 14 synthetic-event selftest fixtures
(T04/T05/T13). Class-level docstring documents the WinAppDriver
limitation and the synthetic-event fallback.

Also bumps the fixture's diagnostic trace ring buffer from 6 to 12 so
post-drag failure dumps show enough pipeline stages to localize a
regression (Press → threshold → BeginTearOff → Tracker.Start → ...).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* test(045): address PR #426 CR feedback

- Move EnsureDpiAware() to ClassInitialize so process-wide DPI context
  is set deterministically before any WinAppDriver session opens
  (Copilot review). Check the return value and log Win32 errors rather
  than silently assuming success.
- Replace broad catch{} blocks with specific exception types — aligns
  with the recent #420/#421 sweep:
    * EnsureDpiAware: catch EntryPointNotFoundException /
      DllNotFoundException (older Windows missing the API).
    * DumpDiagnostics.Read: catch WebDriverException only.
    * StopAppSession driver Quit: catch WebDriverException only.
- WaitForTextAcrossWindows: temporarily zero the implicit wait so the
  100 ms polling cadence actually fires; otherwise the driver's 2 s
  implicit wait per FindElement dominated and a 5 s timeout yielded
  only ~2 polls (Copilot review).
- Drop the stale "E05" reference from the drag-helper comment — only
  E01-E04 exist (Copilot review).

Verified: dotnet test --filter ~DockingTearOffE2ETests → 4/4 green.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: codemonkeychris

tests/Reactor.AppTests.Host/Fixtures/DockingTearOffE2EFixtures.cs

@@ -84,7 +84,7 @@ public override Element Render()
                     updateLastTrace(prev =>
                     {
                         var next = prev.Add(msg);
-                        return next.Count > 6 ? next.RemoveRange(0, next.Count - 6) : next;
+                        return next.Count > 12 ? next.RemoveRange(0, next.Count - 12) : next;
                     });
                 };
                 return () => { DockTabTearOff.DiagnosticSink = previous; };