The 'memory leak' detection in cluster and worker aggregation prevents supporting prometheus and openmetrics in the same process

[jdmarshall]

There are better ways to prevent memory leaks for event listeners in unit tests that don't clobber runtime functionality in the bargain.

[jdmarshall]

The PR I have open for this breaks the per instance lookup tables into file-scoped, and relies on the fact that any parallel requests have distinct IDs which will end up routing the responses to the proper listener.

That could still have a problem with watch commands but is more of an issue when you're actually trying to debug telemetry instead of something else in your application. We could do something with a global variable to track the counter, so that reloads still result in monotonically increasing numbers.