The plan of automatically handle retries on read sessions didn't really pan
out since we don't really control when data is transported over the network
and errors are noticed. Instead we treat reads and writes in the same way, i.e.
throwing a `SessionExpiredException` on all connection failures.

The connection pool is no longer responsible for maintaining the servers
since we must treat READ and WRITE servers differently. Instead added
a separate containers for the different server types.

Instead of two separate procedure calls, one to do routing and one to do
acquisition we merge them into one call containing the same information.