Ignore errors on TCP send to avoid race with incoming disconnect message

THis commit avoids a race condition between an incoming disconnect
message and an attempt to send a new outgoing message. In some cases
the prior implementation would process the TCP close before finishing
processing of messages in the TCP receive buffer, reporting a
BrokenPipeError insted of a clean disconnect.

asyncssh/connection.py

@@ -1413,10 +1413,10 @@ def _send(self, data: bytes) -> None:
         """Send data to the SSH connection"""
 
         if self._transport:
-            if self._transport.is_closing():
-                self._force_close(BrokenPipeError())
-            else:
+            try:
                 self._transport.write(data)
+            except BrokenPipeError:
+                pass
 
     def _send_version(self) -> None:
         """Start the SSH handshake"""