test(tui): relax yolo ask-rule event timeout on windows

[Hmbown]

Summary

• Give model-driven engine-event regression tests a larger timeout on Windows runners.
• Keep the existing 10s timeout on non-Windows platforms.
• Covers the Windows-only full-suite failure seen on main after docs: update ask-rule persistence docs #3642 and on feat(setup): add setup summary wizard step for MCP/skills/plugins overview (#3407) #3643.

Validation

• cargo fmt --all -- --check
• CARGO_TARGET_DIR=/Volumes/VIXinSSD/codewhale-target cargo test -p codewhale-tui --bin codewhale-tui --locked core::engine::tests::yolo_mode_does_not_prompt_for_model_driven_typed_ask_rule -- --nocapture
• CARGO_TARGET_DIR=/Volumes/VIXinSSD/codewhale-target cargo test -p codewhale-tui --bin codewhale-tui --locked core::engine::tests::yolo_mode_does_not_prompt_for_background_shell_safety_floor -- --nocapture

[claude]

Claude finished @Hmbown's task in 46s —— View job

---

Review

• Read changed files
• Analyze diff and surrounding context

---

Summary: Straightforward and correct. The change extracts a model_turn_event_timeout() helper that returns 30 s on Windows and 10 s elsewhere, then applies it to the two async engine-event loop tests that were timing out on Windows CI.

Correctness — no issues found

• cfg!(windows) is the right mechanism here: it's a compile-time gate, so the Windows path has zero overhead on non-Windows builds and there's nothing to go wrong at runtime.
• 30 s is a sensible ceiling for a Windows runner doing shell-spawning I/O inside an async engine loop. It's generous without being unbounded.
• The two call sites (yolo_mode_does_not_prompt_for_model_driven_typed_ask_rule and yolo_mode_does_not_prompt_for_background_shell_safety_floor) are the only places in the file that hit the Duration::from_secs(10) pattern for engine-event polling — confirmed by grep, no remaining hardcoded 10 s timeouts are left behind.
• The other two timeout(Duration::from_secs(2), …) calls at lines 3237/3261 are for unrelated lightweight queries (provider-runtime-status), not shell-execution event loops, so they don't need the same treatment.

Placement — fine

The helper sits between file_ask_rule_engine and the first #[test], a natural spot for shared test utilities. No concerns.

Reuse — nothing to consolidate

Only two call sites exist and they're both updated. No dead code introduced.

Verdict: Ready to merge as-is. Good minimal fix for the Windows CI regression.

---