Replica tracking simplification

[lolski]

Currently, the client keeps track of various aspects of the replicas of a given database: role, term, and 'is prefered secondary' status. The information is stored in the ClusterDatabase class that is quite heavy.

Keeping track of this information on the client is complicated, and we need to minimise it in order to make the client is as thin as possible.

The minimum number of information that can be stored is the IP addresses of all servers. The other information can be queried from the servers as needed.