test(ci): skip two more stinger tests that never park on the Windows runner

[srperens]

Refs #1069

Problem

Main went red on the merge of #1067 (run 38282030983). Two stinger tests outside the Windows skip list failed on the Windows runner with the #1069 symptom: the clip source never parks.

a_clip_that_will_not_play_still_changes_the_program ... FAILED
a_take_that_fails_to_program_leaves_the_old_source_on_air ... FAILED
stinger never ready: None; running false, cued Some(0), parked false, paused true, current_index 0, ...

Change

Add windows to the skip_known_ci_failure call of both tests. a_take_that_fails_to_program_leaves_the_old_source_on_air was already skipped on macOS. a_clip_that_will_not_play_still_changes_the_program gets a Windows-only skip. Linux CI and local runs still run both.

Rejected: skipping the whole stinger suite on Windows. The failure is intermittent per test, and the rule from #1067 is to skip only a test that has been seen failing on that runner.

Evidence

• Both failures come from the Windows job log of run 38282030983 (job 114922147817), with the panic quoted above.

Tests

Ran: cargo fmt --all -- --check locally.
Not run: the tests themselves. The change only adds an early return, and it applies only when STROM_SKIP_KNOWN_CI_FAILURES is set on Windows. CI on this PR shows whether the Windows job goes green.

Out of scope

• The root cause of the park failure, tracked in Stinger clip source sometimes never parks its first clip on Windows #1069.
• In the same run, the Linux job failed on gpu_gbr_10le (vision_mixer_input_format). That is the known software-GL frame-count flake, so it stays unskipped and a rerun is enough.

🤖 Generated with Claude Code

[srperens]

Verdict: Approve. Posted as a comment review because this PR is authored on the account the agent runs under, and GitHub refuses an approval from a PR's own author. The two skips follow #1067's per-runner rule, the failure they cover is in main's own log, and the Windows job at this head is green with both skips taking effect.

Claims

Claim	Verdict	Evidence
Both tests failed on main's Windows runner with the #1069 symptom	CONFIRMED	Run 38282030983, job 114922147817: a_clip_that_will_not_play_still_changes_the_program ... FAILED, a_take_that_fails_to_program_leaves_the_old_source_on_air ... FAILED, stinger never ready: None; running false, cued Some(0), parked false, paused true
The skip applies only on Windows and macOS CI	CONFIRMED	backend/tests/common/mod.rs:65-67 returns false unless STROM_SKIP_KNOWN_CI_FAILURES is set and std::env::consts::OS is listed. .github/workflows/ci.yml:645 and :743 set it only in the two platform jobs
The skip fires on the Windows runner	CONFIRMED	PR run 38360463388, Build (Windows): each test reports ok about 1 ms after the line before it. In main's run 38298278726 at b127864 the same two ran unskipped and took about 13 s

Diagnosis: This is suppression, by design, and the PR says so. The root cause stays in #1069. The main run at b127864 passed both tests unskipped on Windows, which matches #1069's "a different set each time". So the per-test list grows only as failures are seen. After this PR, 7 of stinger_test.rs's 13 tests skip on Windows: backend/tests/stinger_test.rs:663 — &["windows"], plus the six calls listing "windows". The rejected alternative, skipping the whole suite, is recorded in the body. Whether to keep adding tests one at a time or fix #1069 first is a maintainer call, and it does not block this PR.

Radius: LOCAL. One test file, two early returns.

Overlaps: #1069 owns the root cause, and fixing it makes these skips removable. #1055 also changes stinger tests, but in common/stinger.rs and stinger_api_test.rs, not this file. No open issue's design is settled by this.

Tests & CI: run 38360463388 at f867b9f. Check (Linux), both Linux builds, Build (Windows) (ci:windows label), WASM, API Contract and Static JS pass. macOS is skipping. On macOS the first test was already skipped. I ran nothing locally.

Confidence: HIGH