test(runtime-host): validate Windows named-pipe clients

[liugddx]

Summary

• validate Windows named-pipe endpoint naming and lifecycle hooks
• enforce the storage root identity invariant before selecting either endpoint transport
• run transport-neutral two-client Agent Graph, Automation, Daily Review, Deep Research, OAuth, Plan, and Session effect flows on Windows
• replace misleading POSIX UDS skip reasons with the actual remaining Windows lifecycle blockers

Root cause

The runtime host already selected a Windows named pipe, but its endpoint contract and most transport-neutral two-client flows were skipped under a generic POSIX UDS integration label. The Windows branch also returned before validating the 64-hex-character storage root identity required by the POSIX path.

The tests that still fail because of execution-host startup or SQLite shutdown remain skipped with explicit reasons; this PR does not present those lifecycle guarantees as complete.

Validation

• Windows named-pipe and two-client group: 10 pass / 0 fail / 4 skip
• full repository build
• full repository typecheck
• Biome check
• git diff --check

The current full runtime-host Windows baseline remains 578 pass / 116 fail / 25 skip; the failures are existing SQLite, filesystem replacement, and process-lifecycle backlog. All tests enabled by this PR pass in the full run.

Relates to #2142.

[Astro-Han]

Thanks for closing this Windows coverage gap. Reusing the existing transport-neutral two-client tests is the right approach: it validates the real Host/Client contract without creating a parallel Windows suite. Moving the storage root identity invariant before transport selection is also the correct ownership boundary, and the remaining skips now describe concrete lifecycle blockers rather than the transport itself.

I found no actionable issues in the current change. Approved.

Non-blocking follow-up: now that these contracts are known to pass on Windows, it would be useful to promote them into a focused blocking Windows lane later, without waiting for the full Windows baseline to become green.

简体中文

感谢补上这部分 Windows 覆盖。复用现有与 transport 无关的双客户端测试是正确方案：它验证真实 Host/Client 契约，同时没有另外复制一套 Windows 测试。将 storage root identity 不变量移动到 transport 选择之前，职责也正确；剩余 skip 现在描述的是具体 lifecycle blocker，而不是笼统归因于 transport。

当前改动没有发现需要处理的问题，Approve。

非阻塞后续建议：既然这些契约已经确认可以在 Windows 通过，后续可以把它们晋升到一个聚焦的阻塞式 Windows lane，不需要等待整个 Windows baseline 全部转绿。