Problem: boot a cluster with only etcd and controlplane nodes, with no
worker coredns doesn't come up and the cluster-agent can not talk back
to the rancher server url and make the cluster accessible. additionally,
targeting for the shell fails due to taints.
Fix: Add dnspolicy default to the cluster-agent so when it boots it will
use the node's dns settings. Add tolerations to the shell/helmop for
controlplane nodes so the pods will assign and the tunnels will com up.