When a node goes down, it would be kicked from other nodes in the cluster
by global. This process in not instant. This process happens even
if node is alive and could reconnect.

Reconnecting too fast would interfere with prevent_overlapped_partitions logic
though.

We still allow for this node to reconnect, but only during the regular checks.
The change temporary puts the node into unavalable_nodes list though

This allows us to try to reconnect to the remote node
after some period of time after the netsplit.

Enters regular phase once we contact the remote node.

…overy_nodes

Call peer to stop the processes
Use proc_lib to spawn processes for better error reporting in tests

More reliable disconnect_node in tests
Link disco process in tests