Integration tests reliability fixes

[chkeita]

Summary of the Pull Request

Multiple fixes for #1454

• Check if the node is in the don state before reimaging. This gives the node the opportunity to process the StopCommand before we reimage it
• Log node done event as warning when the node is deleted from the database
• Log heartbeat from deleted nodes as warning

Validation Steps Performed

run check-pr

[ranweiler]

Because of way the existing code is written, I find it hard to see how the change to the default behavior of stop_if_complete() interact with various other flows in the system.

Generally: what other system behaviors are determined by that function, and how did we test that certain error cases are still handled correctly?

Specifically: I have a vague concern that there may be error scenarios where we really want to reimage an unresponsive node, but we've now made ourselves rely on its cooperation, and so reimaging will never happen. I don't have an example, but I'd like to think through and test some cases like that.

[chkeita]

@ranweiler

The change i introduced make it so that we don't force the state to done when stopping a task. We rely on the machine to report the state for that.
for the case of unresponsive node, the state is forced to done before reimaging so my change will not prevent the reimage