[thread-cert] handle BrokenPipeError on resetting node in simulator (#13495)

When a simulated node restarts (e.g. during `factoryreset`), the
child process re-executes itself via `execvp`, disconnecting its
UNIX domain socket connection to the virtual-time simulator.

If `VirtualTime._send_message()` attempts to send an event before the
socket disconnection event has been processed by `selectors`,
`sock.send()` raises `BrokenPipeError` or `ConnectionResetError`,
even though the node is marked as temporarily offline in
`_maybeoff_ports` (via `maybeoff=True`).

This commit catches `BrokenPipeError` and `ConnectionResetError` in
`VirtualTime._send_message()`, verifying that the destination port is
present in `_maybeoff_ports` and skipping delivery cleanly.

tests/scripts/thread-cert/simulator.py

@@ -511,7 +511,12 @@ class VirtualTime(BaseSimulator):
         else:
             sock = self.devices[port].get('sock', None)
             if sock:
-                sent = sock.send(message)
+                try:
+                    sent = sock.send(message)
+                except (BrokenPipeError, ConnectionResetError):
+                    assert port in self._maybeoff_ports, f'The node {port} is unexpectedly off'
+                    dbg_print('skip sending message to off node', port)
+                    return
             else:
                 assert port in self._maybeoff_ports, f'The node {port} is unexpectedly off'
                 dbg_print('skip sending message to off node', port)