Increase channel buffer size from network worker to node tasks

[tomaka]

Right now the buffer size is 5 elements. In other words, the worker can send 5 elements to a libp2p tasks, after which it has to go to sleep waiting for the libp2p task to make some space in that queue.

In practice, however, grandpa seems to sometimes send bursts of hundreds of messages, and it probably wouldn't hurt to increase this buffer size to a bit more than 5.

This requires changes in libp2p to make this buffer size configurable, but I'm opening it here to not forget to update Substrate later.

[tomaka]

It seems that this change (#6064) is indeed positive.

We deployed around 11:37.
CPU usage of tasks (red is the network worker):

Size of queue of messages to send out:

These changes indeed seem to indicate that the network worker is often waiting for some node task to be ready, and that during this wait there's an accumulation of messages in the buffers, leading to the network worker taking a long time to process them afterwards.

[tomaka]

I'm going to reopen this, because it seems to be that these channel sizes are still not as large as they should be.

Before blindly increasing the sizes again, we should introduce some way to know when the network worker is waiting for a libp2p node.

[tomaka]

There has also definitely been an improvement to these spikes of networker worker CPU work.

[tomaka]

No longer needed after #6692