test(desktop): scope slash menu refresh observer

[liuxiaocs7]

Summary

Scope the slash-command menu refresh regression watcher to the open menu and the original listbox and Skills group nodes. Wait for each session's real invocable-skills projection request before checking that those nodes remain connected, avoiding unrelated overlay removals and fixed timing delays.

Fixes #3727

Verification

• Targeted stress run: 20/20 passed
• Confirmed the test fails with the fix(desktop): slash picker alternates between two popup sizes after typing /r #2667 regression restored
• Desktop E2E target and isolated reruns passed
• Desktop unit: 1458/1458 passed
• Runtime, Runtime Host, Storage, and Eval suites passed
• Lint, format, Desktop typecheck, and ASF header checks passed

AI use

Select exactly one:

• No generative tool made a substantive contribution
• Generative tooling made a substantive contribution

Tool(s) and scope: OpenAI Codex implemented the test synchronization and observer scoping changes, added the E2E-only preload signal, and assisted with verification.

Checklist

• Tests cover the change and fail without it
• Lint, format, typecheck and the affected suites pass locally

Does this PR entail a change in behavior?

• Yes — described under Summary above
• No

[Astro-Han]

Thanks for the thorough fix! We chose to merge #3737 for this issue — it achieves the same correction (narrowing the observer to close the over-broad window) with zero production surface (1 file, no preload changes), which is minimal for a test-only flake. Your approach with a deterministic waitForInvocableSkillsCall is causally stronger, but we preferred to avoid expanding the preload test seam for this flake. Appreciate the effort — closing as superseded by #3737.

[M4n5ter]

English

I reviewed exact head 7e99237294856c8e1dcbdc1c10ac703966716efc and found no P0–P3 issues.

This keeps the regression test and replaces timing-based observation with causal synchronization: the test first drains the seed Turn/catalog work, then waits for each invocable-skills projection request before asserting that the original menu, listbox, and Skills-group nodes remain connected.

I verified the guard is meaningful by temporarily restoring the old skills: [] transition; the test failed and recorded the Skills group being removed. With this head restored, the target passed 20/20 stress runs, and the complete Desktop E2E suite passed with 65 tests and 1 expected skip. The exact-head hosted test check also completed successfully.

There are no unresolved review threads or stale approvals. Current main has no changed-path overlap with this PR, and the merge tree is clean.

简体中文

我审查了精确提交 7e99237294856c8e1dcbdc1c10ac703966716efc，没有发现 P0–P3 问题。

该修复保留了回归测试，并用因果同步替代基于时间的观察：测试会先排空种子 Turn/catalog 工作，再逐轮等待 invocable-skills 投影请求完成，然后断言原始 menu、listbox 与 Skills group 节点仍保持连接。

我临时恢复了旧的 skills: [] 过渡来验证守卫价值；测试会失败并记录 Skills group 被移除。恢复当前提交后，目标压力测试 20/20 通过，完整 Desktop E2E 为 65 项通过、1 项按预期跳过。精确提交的 Hosted test 也已成功完成。

当前没有未解决的 review thread 或过期 approval。最新 main 与该 PR 没有变更路径重叠，merge tree 干净。