test: fix flaky UpdateTab reconcile flow (Restarting-button race) (#2065)

[atomantic]

Closes #2065

Root cause

The UpdateTab reconcile flow tests intermittently failed CI with:

TestingLibraryElementError: Unable to find role="button" and name "Restarting..."

The component mirrors the updating state into updatingRef via a separate passive effect (useEffect(() => { updatingRef.current = updating }, [updating])), and handleDisconnect's guard reads that ref synchronously:

const handleDisconnect = () => {
  if (!updatingRef.current) return;   // <-- reads ref synchronously
  setTimeout(async () => { ... arm restart polling ... }, 1500);
};

Because runUpdate() calls setUpdating(true) outside any act (after two awaits), the "Reconciling..." button — driven directly by the updating state in JSX — can commit to the DOM and satisfy the test's waitFor(...'Reconciling...') a scheduler tick before the ref-sync passive effect runs. In that window updatingRef.current is still false, so firing disconnect hits the early-return, the 1.5s confirmation timer is never armed, and the button never becomes "Restarting...". This is a pure test-timing race — in production the ref-sync effect flushes long before any socket disconnect, so product behavior is correct.

Fix (test-only)

Flush pending passive effects (await act(async () => {})) before firing disconnect in the shared fireDisconnectAndConfirm helper (used by both failing cases, including the StrictMode one) and in the unmount case's inline block. This syncs updatingRef to the state production always reaches over real network time. No product code changed; all original assertions (restart-polling fallback arming, { silent: true } health check, StrictMode mountedRef survival) are preserved.

Test plan

Ran the file 25x green after the fix (had reproduced the flake ~1 in 8 on clean pre-fix code):

cd client && for i in $(seq 1 25); do npm test -- UpdateTab; done
# 25/25: Tests  5 passed (5)

https://claude.ai/code/session_01TSuyZ5XG9JiijEibeJjL6T