Fixed setup() memory leak

[tadkollar]

Summary

There were a couple of things causing the memory leak in Group._get_scope(). The big one was that _scope_cache was using refs to subsystem instances as dict keys. So when they were dereferenced elsewhere during setup(), the _scope_cache dict still held refs to the old objects because there's no automatic way to clear it out.

Initially, I addressed this by using the id() of the subsystem objects as dict keys, but there was still a small leak. This was because following a setup(), the subsystem instances were different even if the info in _scope_cache was the same. I ended up using the subsystem pathname as the key, and testing on cache hits to make sure it still referred to the same object.

I added a test that runs setup() 10 times and then 100 times and checks for differences in memory usage afterwards. If the difference is more than 10k, it fails.

Related Issues

• Resolves Possible memory leak in compute totals #1746