Up to now consensus instances used the main channel to communicate with
the background network worker. This lead to a race condition when
sending a local collation and dropping the router before driving the
send local collation future until it is finished. This pr changes the
communication between worker and the instances to use their own
channels. This has the advantage that we don't need an extra
`DropConsensusNetworking` message as the network is dropped
automatically when the last sender is dropped.