Remove node self-deletion behavior on GCP and DO

[dghubble]

• Node self-deletion is now only performed on AWS clusters
• Differentiating an unhealthy/NotRead node from a node that has shutdown (temporary reboot or permanent preemption) is tricky. To cut down on noisy alerts, we've previously favored having nodes delete themselves when shutdown gracefully.
• Though node self-deletion is useful, we may now be able to remove this behavior on some platforms to alert in more cases where an alert is warranted
• On Digital Ocean, there is no managed instance group / ASG to replace a deleted node. Losing a node for a long enough period (reboots are fine) probably merits an alert that's not sent today with the self-deletion design. As a tradeoff, admins performing a terraform scale-down must use kubectl to manually delete nodes that are removed.
• On Google Cloud, node self-deletion prevents alerting when a significant number of nodes are preempted. Fortunately, Google Cloud uses consistent naming of a node between preemptions so reboots and the daily preemption shouldn't trigger an alert. A node that's preempted and doesn't get replaced for a long enough period does merit an alert. As a tradeoff, admins performing a terraform scale-down must use kubectl to manually delete nodes that are removed.
• On AWS, spot instances lack Google Cloud's consistent naming feature. We must keep using the node self-deletion behavior to avoid preempted spot instances from accumulating and causing an alert (preemption and replacement is normal and fine). Self-deletion does mean in cases where many spot workers are preempted, no alert will be sent (undesired).

[dghubble]

I just can't really overlook the difficulty this presents for true autoscaling cases, either offered by the cloud provider or via a privileged agent.