Currently in the rack unlock code (as of #487), shares are retrieved serially, and we always try to contact all remote sleds to retrieve them when we fail to get a quorum. We really only want to contact sleds that haven't responded when we retry, and also make retrievals work in parallel.