Ensure we can upgrade k8 version - in place

[lewismarshall]

Note, there maybe no steps required here before going live as keto could just orchestrate new master / compute nodes.

This ticket is more an issue of:

1. handling upgrades in place without restarting rebuilding nodes
2. upgrading k8 objects

At the moment keto-k8 shares the manifests using etcd to provide "pre-self hosted" multi-master. At the moment, before self hosted clusters, we would need to introduce k8 version detection on master start (re-share using etcd). - Only some cluster unique pki resources are shared.

This will NOT upgrade essential deployments (proxy or dns) or node configuration which will come after this has been completed:

Steps required:

• Detect k8 version change from cloud provider
• Re-write the master manifests and re-share between the masters
• Upgrade other k8 api objects by using later version of kubeadm when steps become idempotent, see Make all kubeadm post-install tasks idempotent kubernetes/kubeadm#288