* Kubelets can hit the ContainerManager Delegation issue and fail
to start (noted in 72c94f1c6). Its unclear why this occurs only
to some Kubelets (possibly an ordering concern)
* QoS cgroups remain a goal
* When a controller node is affected, bootstrapping fails, which
makes other development harder. Temporarily disable QoS on
controllers only. This should safeguard bring-up and hopefully
still allow the issue to occur on some workers for debugging
* Use the newly minted "Fedora CoreOS Preview" AMI
* Remove iscsi, kubelet.path activation, and kubeconfig
distribution
* As usual, bare-metal efforts make cloud provider ports
much easier
