[Task] Windows advisory: session loop tests time out at 3000ms

[Astro-Han]

Symptom

The Windows advisory workflow failed on dev sha 3c15b376b after PR #552 merged.

Run: https://github.com/Astro-Han/pawwork/actions/runs/25666988353

Failing job:

• unit-windows-opencode-session

Failing tests:

• cancel interrupts loop and resolves with an assistant message timed out after 3000ms
• concurrent loop callers all receive same error result timed out after 3000ms

unit-windows-opencode-server-tools was cancelled after the session shard failed. unit-windows-app, unit-windows-desktop, and unit-windows-opencode-config-project passed.

Initial RCA

This does not appear related to PR #552. PR #552 only changed packages/app child-store/sidebar timestamp cache reads, while the failure is in packages/opencode session loop/cancel tests.

This matches the same advisory-only Windows timing pattern as #546 and the earlier PR #543 fix: tests with a hard 3000ms timeout can exceed the budget on Windows runners even when the behavior is otherwise correct.

Candidate Fix

Use the established slowIOTimeout test constant for these Windows-sensitive session loop tests, following the PR #543 pattern.

Priority

P3. Advisory-only test debt. macOS/Linux PR gates are unaffected.

Related

• [Task] bootstrapDirectory > marks session status as error test fails on Windows runner (advisory) #546
• PR test(opencode): bump prompt-effect busy/idle timeout for windows runner #543