Tempest cleanup function improved, expanded, and made more useful

Current tempest cleanup uses a saved state to know what not to delete. This approach works in most situations when working alone on a single node, but is unusable if used in a multi-node build. One example issue would be if tenant A issued a saved state, and then tenant B adds a new resource. Tenant A runs cleanup and tenant B's resource is deleted. We need a more dynamic solution to cleaning up test resources and orphans. Please join my discussion on etherpad in order to propose and brainstorm solutions. https://etherpad.openstack.org/p/roomba