Membership tombstone timeout and "gc" #134
Description
When we down nodes, we have to store their associations as tombstones for a while, in case it'd still somehow reach back to us and rather than rejecting it we'd reconnect it after it was downed.
Such tombstones accumulate over time, so we have to clean them up.
A timeout of e.g. one day or "many hours" hours should be good here. The task can run quite rarely.
Tombstones are currently in ClusterShell._associationTombstones, we'd store .tombstone(node, tombstoneDeadline) there and remove any that are to be removed.