Cluster: Harsh tests for "down node, yet it extends an handshake back"

Our .down-ing moves a UniqueNode into the Tombstones set in the ClusterShell.

When we get new handshakes we MUST ensure that the node wasn't already tombstoned -- this would make us reopen a connection to a node we downed, breaking our promises "to never talk to it again".

Start by locating ClusterShell, the .down-ing path for UniqueNode, and the Tombstones set, then trace how new handshakes are accepted. Add focused tests for a downed node sending a handshake and verify that it remains tombstoned and no connection is reopened.

Written by the indexing model from the issue text.