This PR appears to have caused a large performance regression when listing large B2 buckets. I'm confirming and creating an issue now.

I'm not sure that it's possible to successfully implement this functionality. Consider a bucket with millions of objects with keys of the form folder00000001/foo, folder00000002/foo, folder00000003/foo, etc. This is not a particularly contrived example - analytical data can have this kind of structure. Every 'folder name' will need to be inspected, via a B2 API call, to determine whether it should be shown.