fix(test): give the last three dashboard integrations a real timeout

Three integration tests fail on CI, all the same way:

(fail) dashboard application onboarding integration ... [5000.72ms]
(fail) dashboard configuration integration ...         [5000.59ms]
(fail) dashboard source connection integration ...     [5000.58ms]
  ^ this test timed out after 5000ms.

Why

They pass standalone at ~2s each. The cost is not the test body. Starting the dashboard spawns a real stx build --no-cache of the cockpit. Measured on the source connection test:

mkdtemp            2ms
hashPassword      51ms
saveUsers          0ms
startDashboard  1756ms      <-- buildLiveUi spawns the stx build
POST /api/login  155ms

Ten dashboard integration files each pay that. Under the full 313-file run the contention pushes these three past the 5000ms default.

They are also the only three dashboard integration tests with no explicit timeout. Every sibling already has one:

dashboard-security documents this exact failure in a comment: "runs in ~3s locally but measured 5013ms on CI - flaking the whole pipeline on a 13ms margin." This applies that same fix to the three that were missed.

What I tried first

Skipping the live UI build in tests, via a liveUi option defaulting off under NODE_ENV=test, matching the existing queueWorker and spendLoop flags. That does not work: in a source checkout the live build is the only UI root there is. The packaged UI needs bun run build, so with the build disabled startLocalDashboardServer throws ts-cloud dashboard UI not found and all ten fail instead of three. Reverted.

Making the boot cheaper still looks worthwhile, since ~1.7s per dashboard boot is most of the runtime of ten test files. But that is a change to how the dashboard boots, not a test fix, so it does not belong in the same change as unbreaking CI.

Verification

Full suite after the change: 4117 pass, 0 fail (12 skip, 2 todo).