Refactor Marked for Termination feature

[lrgar]

On this PR I'm refactoring when Marked for Termination events are going to be sent: rather than sending them when a node goes cordoned or deleted, we'll now only send them when deleted.

The reason is that a node being cordoned doesn't affect DaemonSets such as the OneAgent; they are still running, so the timings can be wrong.

So the idea here is to keep a cache of seen nodes in a ConfigMap on the dynatrace namespace (called it dynatrace-node-cache. I've modified the Nodes controller to run periodically every 5 minutes to check the current state and send events if a node on the cache doesn't exist anymore. I've also added the logic to recognize deletions from the Dynatrace API.

As for additional notes,

• The usual Controller pattern from the the controller-runtime doesn't fit our purposes, so the Node Controller is now a "Runnable".
• The cache ConfigMap is owned by the Operator Deployment, so will be removed on uninstallations, etc.
  • And to set this ownership, I need to query for ReplicaSet and Deployment objects, so I've added the permissions to the Roles.
• In case of failures, unlike the standard Controller logic, I'm just logging the error and will try again on the next full reconciliation (so in 5 minutes.)