CI: flaky Textual TUI tests on Windows (NoMatches on #status-bar/#history-table) block PRs

[elkaix]

Symptom

The test (windows-latest, 3.12) leg of ci.yml intermittently fails with:

textual.worker.WorkerFailed: Worker raised exception: NoMatches("No nodes match '#status-bar' on Screen(id='_default')")

The failing TUI test varies between runs (observed: tests/test_tui.py::test_run_cleans_the_selected_file_and_reinspects, tests/test_tui.py::test_clean_now_runs_the_plan_and_shows_the_run_pane, tests/test_layer_b_end_to_end.py::test_tui_run_performs_a_live_rewrite) but the root error is always the same: a worker queries a selector before the widget is mounted.

Evidence (last ~1 hour, unrelated diffs)

• Run 34542718223 (main, gradio bump) — failed this way at 23:34
• Run 34542483061 (tokenizers dependabot) — failed at 23:30, passed at 23:46 with no code change
• Run 34546593588 (naming-fix PR fix(naming): correct watermarks-remover to watermark-remover in service strings; align README SynthID row with CONTEXT #49) — failed twice (two different TUI tests), passed on third attempt

Since ci-green is a required check, this flake randomly blocks PR merges.

Suggested direction

• Assert/wait for widget mount before workers query selectors (e.g. pilot + explicit mount wait, or App.run_test() with wait_for_animation=False + polling query_exclusive/retry)
• Or mark the TUI tests pytest.mark.flaky(reruns=N) on Windows as a stopgap
• Or serialize the TUI tests (some Textual flakes are load-order related)