ClusterAPI deployment failing with occm due to incorrect node selection

[millingw]

Following this commit on 2nd July, bd8ccfe, new cluster deployment via ClusterAPI on OpenStack is failing to successfully deploy the control plane. We think this is because the node selector condition node-role.kubernetes.io/control-plane: "true" is never satisfied (always an empty string ""), and the openstack controller manager pods are never scheduled. There was a related merge request https://github.com/kubernetes/cloud-provider-openstack/pull/2929/files that looks like was intended to go with this change and would fix the behaviour, but has not been merged in. A workaround is to explicitly set the nodeSelector value in https://github.com/kubernetes/cloud-provider-openstack/blob/master/charts/openstack-cloud-controller-manager/values.yaml#L52 before deployment, but it'd obviously be better if the out-of-the-box behaviour was correct.