python: on ProtocolChannel close, close the transport

Right now ProtocolChannel basically ignores "close" from the user.  The
correct this to do in this case is to ask the transport to close.  The
"close" message will be sent back to the user once we see our
.connection_lost() method called by the transport.

Doing this means that we no longer need to worry about killing the
subprocess from the stream channel — it happens as part of .close() on
the SubprocessTransport.

src/cockpit/channel.py

@@ -336,6 +336,10 @@ def do_done(self) -> None:
         if self._transport.can_write_eof():
             self._transport.write_eof()
 
+    def do_close(self) -> None:
+        if self._transport is not None:
+            self._transport.close()
+
     def data_received(self, data: bytes) -> None:
         assert self._transport is not None
         if not self.send_data(data):

src/cockpit/channels/stream.py

@@ -107,19 +107,6 @@ def do_options(self, options):
         if window is not None:
             self._transport.set_window_size(**window)
 
-    def do_close(self):
-        assert isinstance(self._transport, SubprocessTransport)
-        pid = self._transport.get_pid()
-        # avoid calling .terminate(), as that will try to read the process' exit code and race with
-        # asyncio's ChildWatcher (which also wants to wait() the process); the cockpit.spawn() API
-        # already ensures that the process is valid before even calling do_close.
-        try:
-            os.kill(pid, signal.SIGTERM)
-            logger.debug('Received close signal from peer, SIGTERMed process %i', pid)
-        except ProcessLookupError:
-            # already gone? fine!
-            logger.debug('Received close signal from peer, but process %i is already gone', pid)
-
     async def create_transport(self, loop: asyncio.AbstractEventLoop, options: Dict[str, object]) -> SubprocessTransport:
         args = options['spawn']