tests to demonstrate async-to-async unloadAll(<type>) issue (#9151)

[runspired]

these tests demonstrate the unloadAll() edge case described in #9151

After debugging these tests, I've reached the conclusion this issue cannot be reasonably addressed without a major rewrite of our GC behavior. The unload APIs have always been a poor-man's escape hatch with tons of caveats, and I'm inclined to believe the lift needed to fix this issue is large enough that its not the right time to pursue it unless someone just happens to really want a fun graph traversal optimization problem to hack on.

At some point we're going to want to remove the unload APIs anyway, either as part of exploring #8162 or whenever we land support for store forking.