etcd_discovery_url is causing terraform to reprovision instances when it doesn't need to

[tayzlor]

Steps to reproduce -

• Provision initial cluster
• Increase number of slaves (+1)
• Run terraform plan - on this step it will change the cloud config for every instance in the cluster (because we are using the size variable, which is taking into account ALL machines in the cluster)

At the moment we are using a "small cluster" architecture for etcd (see https://coreos.com/os/docs/latest/cluster-architectures.html)

We should move to the "Production cluster" setup running the etcd central services only on the "master" nodes. This would mean we'd have a much more static setup (with either 3 or 5 masters) and allow the slaves to scale as they wish and join the discovery URL for the central etcd cluster. As a result size shouldn't be changing too frequently (unless the number of masters is changed - which is fine), which shouldn't cause terraform to taint resources as much circumventing this issue.