Memcached should be changed to be almost the same as RESP. Currently it uses a CompletionStage chain to order events, but this should be changed to cancel auto read on the netty channel instead like RESP.

Also we need to add a check when responding to a command that it disables auto read when the client cannot keep up.