Recursion error on "fake" s3 directories

[pjbull]

This was brought up by @remi-braun in #148.

Sometimes, listing a directory with the S3Client results in a recursion error. This was seen on Ceph where "faked" S3 directories are more common.

@GeorgeSabu found that this is likely the same root cause as what was fixed in #190. We assume that all the "content" items from the API are files, but they can be directories as well. The way to detect which is a directory is to see if the size is 0, as is implemented here:
ElucidataInc@9c11c6a

[GeorgeSabu]

@pjbull I am interested in contributing, here is the PR.

[jayqi]

@pjbull per my comment here, I think that the root cause for this issue may be misidentified. I think it might actually be the issue with download_to + iterdir that was identified in #204, and in that case it would be fixed by #202.

[pjbull]

Between #202 and #302, I think this should be fixed now. Closing unless we see this again.