Flaky test: test_hung_scan_tools_are_unavailable fails under load #136

What: tests/test_health.py test_hung_scan_tools_are_unavailable passes reliably regardless of machine load.
Why: It relies on wall-clock timing (a /bin/sleep 0.3 stub against a patched SCAN_TIMEOUT_SECONDS = 0.05) and failed 5 of 5 runs after the full suite under load, on both main 1fbbb0f and PR #134's head.
So what: Small test fix; whoever picks it up.

Evidence

• Muse Spark 1.3 review of PR fix: treat pending gh pr checks as pending (#132) #134 (2026-10-19, run on cavallo): full suite 860 tests, 1 failure in this test on head and on base; it passed once on base by timing luck. CI proof passed, so it is load-dependent.
• tests/test_health.py around lines 115-116.

Elon record

• Requirements and who asked: reliable proof runs for every PR (operations verification); found during the 2026-10-19 review follow-up Luke requested.
• Deleted: changing health.py behavior; only the test's dependence on timing is in scope.
• Bottleneck: one timing-based assertion.
• Checked myself: the test source lines above; the reviewer's reported reruns (not reproduced by me).

Acceptance criteria

• The test proves the hung-tool path without depending on scheduler timing (for example a stub that never exits, or an injected clock).
• It passes repeatedly under parallel load.

Non-goals

• Changing health scan semantics.

Blockers

None.

Required proof

• 20 consecutive full-suite runs, or the test in a loop under load, all passing; independent review.