db replicator race leaves temporary database in tmp dir

There's probably more than one reason a remote container-replicator might fail to call the REPLICATE RPC after rsyncing a database over (either the ReplicatorRpc complete_rsync or rsync_then_merge) which could cause a temporary db to stack up (and waste disk space) in the temporary dir.

But one common way seems to be a race when two (or more) remote container-replicators are both trying to complete_rsync a remote db onto a new node after rebalance.

If the database is large and the network busy - it's not uncommon to hit such a wide race.

When it does then the loser will miss some cleanup code:

https://github.com/openstack/swift/blob/6e893e228840bc42cfd13546245438832bc2bb46/swift/common/db_replicator.py#L820

While it's probably reasonable to avoid some sort of sync/merge and return the 404 error - before doing so the local container server should cleanup the temporary db which the remote is trying to tell us about.

Otherwise it *will* get reaped if it's older than a reclaim age (lp bug #1691565)