Failed to recover when node sending /localnodes goes down

As described in the https://scylladb.atlassian.net/browse/SCT-370
Alternator loadbalancing library fails to fetch the nodes from the one instance when that instance is down and others are responding. Looks like the library always uses the supplied the IP address to fetch others from /localnodes, but when that node goes down it fails the whole test util.
Suggested fix:

1. Store all other nodes and use them as fallback if the primary supplied node is not available
2. Allow multiple nodes to be supplied to the library and then you be used in round robin fashion for fetching /localnodes and when one goes down it removes it from the list.
3. Combined 1 and 2. Supply multiple and when nodes are discovered it, they also become the fallback for running /localnodes