discover state of existing clusters

[joel-bluedata]

handleStatusGen uncritically accepts any already-created cluster CR that KD sees for the first time and treats it as a stable cluster. Instead, we should do some k8s interrogation, to deal with the case that a previous instance of KubeDirector crashed after making changes but before the cluster CR status could be updated.

(If we do this we can also execute KD in the container in a way so that it is just restarted whenever it crashes.)

[joel-bluedata]

If we do issue #204 and it works as expected then arguably we don't need this, as no one else should be able to edit the resource while KD is down.