kubernetes master: single-point-of-failure for prometheus

[aledsage]

In kubernetes blueprint, one 'kubernetes master' looks like a single point of failure of prometheus etc.

In kubernetes/catalog/kubernetes/kubernetes.bom, the kubernetes-master cluster has a different spec for the first versus subsequent members of the cluster. The difference is that the 'firstmemberspec' also has adds the 'calico pool', 'calico policy controller pod', 'kubernetes dashboard pod' and 'prometheus pod'.

The PR #447 works around this in the QA tests - it ensures we kill a master that is not running prometheus, because we subsequently assert that prometheus is available.