Small bugfix in connection monitor

The connection monitor is a thread we spawn for each RPC; it's job is to
monitor the connection and close the RPC if the entire connection is closed. In
order to prevent memory leaks, the thread also terminates when the RPC is
closed independent of the connection. However, we were waiting for termination
of the _outbound_ thread to determine if the RPC is closed; it's better to wait
for termination of the _inbound_ thread, because the outbound thread is
terminated when the client sends its final message, but at that point we might
still be waiting on further messages from the server.

This doesn't matter too much, but becomes important in the next commit.

grapesy/src/Network/GRPC/Client/Call.hs

@@ -160,7 +160,7 @@ startRPC conn _ callParams = do
     _ <- forkLabelled "grapesy:monitorConnection" $ do
       status <- atomically $ do
           (Left <$> Thread.waitForNormalOrAbnormalThreadTermination
-                      (Session.channelOutbound channel))
+                      (Session.channelInbound channel))
         `orElse`
           (Right <$> readTMVar connClosed)
       case status of