Updated rings were only pushed to the first controller node, resulting
in a data movement loop: controller 1 pushed data to OpenShift and
removed it locally, while controller 2 and 3 were detecting missing data
on controller 1 (because they were still using old rings) and
replicating back to controller 1.

Update both the docs and the test role to distribute rings to all three
controller nodes and restart Swift services on each.

Also add the -ignore_readdir_race option to the find command that checks
for remaining .data/.db files on the standalone node. The replicator may
remove empty directories while find is iterating, causing spurious
errors.

JIRA: OSPCIX-1318

Signed-off-by: Christian Schwede <cschwede@redhat.com>