Flake batch 22: inspect gate wait, subc-core download retry, standing health contention

Merged
AF
4d ago
+48-11

Test and CI fixes only.

  • commands::inspect::deferred_terminal_tests::standalone_shutdown_cancels_detached_inspect_before_phase_deadline gave the detached inspect 1 s to reach its body gate, and a loaded Windows runner took longer. The gate holds the run until released, so it now allows 30 s. The shutdown-cancellation check keeps its own bound.
  • scripts/fetch-subc-core.sh made one attempt to download the pinned subc-core release, so a brief api.github.com connection failure failed main's macOS conformance job before any test ran. It now retries up to four times with 5, 15 and 45 s waits, and the pinned checksum still verifies the result. Tested with a stand-in gh that fails twice and then succeeds, and with one that always fails, which stops after four attempts.
  • subc::health::tests::standing_health_names_resolved_path_drift_without_re_recording read the health report once. Under load the report says the scheduler is busy and falls back to its last census, which may not list the new root yet. The test now retries while contended, like the other health tests, and every assertion is unchanged.

Verified on the build server: the deferred inspect tests and the standing health tests three times each, plus the Windows target check.