Currently the single topology stores its seed address, which is subsequently used for deciding whether a host should be added to a cluster. Other topologies do not store any addresses.

The address storage is confusing because the address stored is a seed address rather than the current address (for example, seed address may be missing the port number).

As part of spec compliance as well as improved diagnostics, topology should take the list of addresses off its cluster. Topologies are tied to clusters hence this does not create additional dependencies. This change also makes topologies more sensible in that their data reflects their current state.

We still need to store the seeds, mostly for transitioning from unknown topology to other topologies, and this job will be handled by cluster for the time being.