Fix for resource leak that occurs when parent scope is closed during resource acquisition

[mpilquist]

This needs a lot more work but committing what I've found so far. When the resource acquisition completed in CompileScope#acquireResource, the wrapper ScopedResource instance was getting registered on the containing scope, but the containing scope's state was already closed. Not sure why the test I added yesterday passed though.

[mpilquist]

Okay, figured it out. The fix from yesterday was only partial. When the resource acquisition completed, the associated ScopedResource was registered with the current scope, which was already closed. However, that scope was still the current scope at stream termination and hence was closed again, freeing the resource. Adding a subsequent bracket resulted in changing the current scope, and hence there was no double closure.

In this PR, we correctly prevent any resources from being registered on a closed scope. This causes the original test to fail as well as the new variant. We fix by invoking the finalizer if registration fails.

I'd like to change these internals a bit -- e.g., why does a closed scope even have the ability to reference resources -- but I'll save the more drastic changes for develop branch.

[SystemFw]

nice work!

I'd like to change these internals a bit -- e.g., why does a closed scope even have the ability to reference resources -- but I'll save the more drastic changes for develop branch.

👍 👍