ci-fix: transient failure on 39689909129 #1354

CI run 39689909129 failed in one test on Standalone Platform Check (macos-latest): agent-browser host viewer > watches a daemon it cannot prove live beside a daemon of the same session name in its own. Every other job passed, including the Windows leg running the same test file. I'm classifying it as a runner-timing flake, with no fix PR.

• What failed: daemon.connected() (agent-browser-host.test.ts#L706) is a vi.waitFor with vitest's default 1000 ms timeout. It waits for the host's viewer to open its WebSocket onto the fake daemon stream. The test failed after 1036 ms with expected undefined to be defined, so the socket never opened within that 1000 ms window.
• Why it isn't the commit: Simplify the Windows/WSL fixes from #1343, #1344, #1347 and #1348 #1353 changed only how run and readStreamPort put --headed into CLI argv. viewStream (agent-browser-host.ts#L737) spawns nothing on this path. It runs liveDaemon (two state-file reads and one loopback port probe), then connects.
• Timing: the same test took 419 ms on macOS in the previous main run 39684387868. That includes its own 300 ms sleep, so the connect normally takes about 100 ms. In this run the neighbouring case took 513 ms and the whole file 23.5 s, against 21.9 s before, which points to a slow runner.
• History: this test does not appear in any of the 60 most recent failed CI runs (back to 2026-10-19). This is its first failure in that window.

If it recurs, the next step is to give fakeServer().connected() an explicit timeout matching the host's own 5 s STREAM_CONNECT_TIMEOUT_MS, rather than relying on vitest's 1 s default for a connection that crosses several hops.