Let the notify test observe the seam instead of the clock

[xmap]

Sibling of #603: a wall-clock assumption that holds serially and not under -n 4. This one had two, and the first was silently unsound.

The problem

wait_task = asyncio.create_task(wakeup.wait(5.0))
await asyncio.sleep(0.05)   # hope add_listener has finished
await _send_notify(db_pool, "wakeup-test")
await asyncio.wait_for(wait_task, timeout=2.0)
assert elapsed < 1.0        # a clock standing in for "the notify woke it"

If the 50ms sleep is too short, the NOTIFY is delivered to nobody and dropped. wait() then falls through to its own timeout, and the test fails on exactly the condition it exists to rule out. The old comment named that race and slept anyway.

The elapsed < 1.0 assertion is the same species: a timing proxy for a semantic claim.

The fix

Both replaced with signals the code already exposes:

was	now
sleep(0.05) and hope	wait on the _listening flag _ensure_listening sets
assert elapsed < 1.0	assert wakeup._event.is_set()

wait() clears _event before waiting; _on_notify sets it. Set means the notify woke it, clear means the timeout did — which is the claim the test's name makes, now asserted rather than inferred.

The remaining bounds are hang guards, not timing assertions. At 20s they sit far above any plausible contention delay and far below the 60s per-test cap, so a genuine hang still fails as a hang.

Verification

Two full runs under -n 4 — the configuration where it failed: 47,331 passed each. Also 3/3 serial and 3/3 under targeted contention.

🤖 Generated with Claude Code

[github-actions]

Coverage report

This PR does not seem to contain any modification to coverable code.