Support force_nodes for baremetal provisioning

With the baremetal driver, force_hosts is not sufficient since the compute host is actually controlling other physical nodes. We need a way to force which baremetal node is provisioned.

The current implementation of force_hosts is done by passing --availability_zone az:host. It seems appropriate to extend this to also support az:host:node. This will require a small change in nova/compute/api.py: _handle_availability_zone and _validate_and_provision_instance to parse host:node. It will probably require much larger changes in the host_manager, filters, and elsewhere to handle the inclusion of nodes.

However, this approach requires that the operator knows which baremetal compute host controls the desired baremetal node, and I would like to avoid that. In order to do that, nodes need to be identified by uuid rather than host:number.