Refactor + separately test the instance deletion logic in task scheduling

[isoos]

• Refactored the instance deletion cycle similarly to the package scan logic in Refactor task loop: scanning for package updated timestamps and updating tracking state. #9082.
• Added tests for the cycle's typical use cases.
• Added a small (and probably not really necessary) wait for the delete operations to be completed. Previously we have completely ignored the pending deletes in the scheduled microtasks, and while this would not completely block on their completion, it makes an attempt to wait for these.
• Added a minimal event tracking to the task backend/scheduler, to make it easier to expose the internals and also to fix the test. I've added a test assertion to explicitly see the timeline of creation / deletions, it reveals that the tests' previous assumptions were not correct, and we are scheduling the cloud instances before the previous one gets deleted.