test: TestRunSurvivesARingWaitEBUSY884/submit-and-wait-timeout fails its apparatus guard in Coverage's -race -covermode=atomic build (9 of ~264 runs) #1107

TestRunSurvivesARingWaitEBUSY884/submit-and-wait-timeout (internal/engine/iouring) fails its own apparatus guard in the Coverage workflow's -race -covermode=atomic build: 9 failures in about 264 Coverage runs that reached the io_uring step (3 of them on main: runs 39068951624, 39334596754, 39460416507), never in the Unit workflow.

Mechanism (reproduced): the arm's 32 unpaced clients keep the loop so busy that it never takes the timed wait the arm exists to exercise, so the test's own guard fires. With a race plus atomic-cover build, one worker, --cpus 4 and no spinners it fails 31 of 40 locally. The clients should pause after each answer in the timed arm; with a 20 ms pause it is 0 of 200 at --cpus 4 (95% upper bound 1.5%) and the full test passes 30 of 30 at --cpus 1 with 3 spinners and 60 of 60 at --cpus 2 with 2 spinners.

Fix: celeris draft PR #1106 (test-only; Refs #N only). Found while triaging the CI flakes of #890 and PR #1105's attempt 1.