Removing a openstack-hypervisor unit does not clean up the resources correctly.

[Pedro Castillo]

When removing an `openstack-hypervisor` unit from a machine, the compute service and hypervisor are not removed from the OpenStack API. Additionally, any VMs running on that node are not removed from the API either.

Given these variables:
```
export JUJU_MODEL=openstack-machines
MACHINE=6
HYPERVISOR_UNIT=$(juju status $MACHINE | awk '/openstack-hypervisor\// {print $1}')
MACHINE_FQDN=$(juju ssh $MACHINE hostname --fqdn | tr -d '\r')
```

Removing the unit will require manual cleanup.
```
juju remove-unit $HYPERVISOR_UNIT
```

This is the manual cleanup required:
```
# Delete VMs from the API.
SERVERS=$(openstack server list --all --host $MACHINE_FQDN -f value -c ID)
for server in $SERVERS; do
    openstack server delete $server
done

# Delete VMs in the machine through Virsh.
juju ssh $MACHINE '
  sudo openstack-hypervisor.virsh list --all --uuid | while read vmid; do
    for cmd in destroy undefine; do
      sudo openstack-hypervisor.virsh $cmd $vmid
    done
  done
'

# Remove the compute service.
COMPUTE_SERVICE=$(openstack compute service list --host $MACHINE_FQDN -f value | awk '$2=="nova-compute" {print $1}')
openstack compute service delete $COMPUTE_SERVICE
```