[bahdotsh]

Main's CI is red and #670 cannot go green, and macos-14 is retired on November 2. This carries #670's commit plus the fixes for both, so merging it also lands #670.

Main (Python Wheel (windows-x86_64), run 41192023539): a timing flake. test_a_slow_call_leaves_the_loop_ticking_and_other_calls_waiting assumed a 1 s stall outlasted its setup; on a slow runner b waited only 0.25 s. The stall is now a gate the test releases, so every step runs inside it. Without the call lock the rewritten test fails.

#670 (Python Wheel (macos-arm64) on macos-15): two separate causes, measured with a probe on both images.

	macos-14	macos-15
gethostbyaddr("127.0.0.1")	2 ms	35 s
zeroconf self-discovery, IPVersion.All	found	not found in 10 s
zeroconf self-discovery, IPVersion.V4Only	found	found

• Five http_front/test_demo.py tests: HTTPServer.server_bind calls getfqdn, so the demo provider sat in reverse DNS past every test timeout. It now skips the lookup, whose result it never read.
• test_a_host_advertised_over_dns_sd_is_found_and_proved: the test now skips when a zeroconf-only probe cannot see its own advertisement. The probe does not touch PeerStreamManager, so a fault in the manager still fails the test.

Not proved: discovery uses IPVersion.All in peer_stream_manager.py and dnssd_bridge.py. It works on macOS 26 hardware, but the cause on the runner was not isolated, so a real macOS 15 machine is unverified, and DNS-SD discovery has no CI coverage on that image.

Update from the green run (682c673): the macOS wheel job reported 1148 passed and 4 skipped, the same 4 as before, so the DNS-SD test ran and passed on macos-15 this time and the skip did not fire. The zeroconf failure therefore varies between runs of that image rather than holding on every one; the table above is one probe, and the skip exists for the runs where it recurs.