Discovery is slow for chains added later

[tomaka]

The networking service periodically starts a discovery process, where it asks a random peer for the peers it knows.
If multiple chains have been added to the networking service, it does this at the same time for each chain.

When the networking service is first created, the discovery fires very quickly in order to quickly fill the peers book. Over time, the delay increases.

This was designed at the time when the networking service could only serve a single chain.

Unfortunately, if a chain is added later, the discovery will already be slow, and the peers book of this chain will be slow to fill up.

[tomaka]

While a good solution might be to slow down discovery once our address book is populated, in practice we can't do that as we have no idea of the size of the network. If a network has only 10 nodes in total, the discovery should slow down as well, even though the address book will never go past 10 entries.

I think that the best approach is simply the existing approach of an interval that increases automatically, except that it should be done per individual chain rather than globally.