Edge runs semi-regularly failing, on wptserve shutdown, after issues near start-up

[gsnedders]

On results_edge_dev/results_edge_stable runs, a serve_forever thread occasionally dies ~1s into the run with a spurious WinError 997 on the shutdown-notification socketpair (from 258.txt):

Exception in thread Thread-1 (serve_forever):
Traceback (most recent call last):
  File "...\Lib\threading.py", line 1082, in _bootstrap_inner
    self._context.run(self.run)
  File "...\Lib\threading.py", line 1024, in run
    self._target(*self._args, **self._kwargs)
  File "...\tools\wptserve\wptserve\server.py", line 280, in serve_forever
    shutdown_read_sock.recv(1)
OSError: [WinError 997] Overlapped I/O operation is in progress

Nothing has called shutdown() at this point — yet the select() loop reports the shutdown_read_sock as readable. serve_forever's finally block closes _shutdown_write_sock, so when shutdown() finally does run at real teardown, its send() hits the already-closed socket and throws, which serve.py logs as CRITICAL:

Process http-local on port 8002:
58:11.02 wptserve CRITICAL Traceback (most recent call last):
  File "...\tools\serve\serve.py", line 1062, in create_daemon
    self.daemon.stop()
  File "...\tools\wptserve\wptserve\server.py", line 995, in stop
    self.httpd.shutdown()
Traceback (most recent call last):
  File "...\tools\wptserve\wptserve\server.py", line 298, in shutdown
    self._shutdown_write_sock.send(b'x')
OSError: [WinError 10038] An operation was attempted on something that is not a socket

Any CRITICAL log line makes wptrunner exit non-zero (exit code 64), so a run that otherwise completed fine still fails outright.