[Bug] Deleting "fake" directories

[MileanCo]

When deleting fake directories that tools like Hadoop create, this tool ends up deleting the entire directory even if not empty... See this https://issues.apache.org/jira/browse/HADOOP-13230

For example, if I have:
test_run123/foo.bar

and the hadoop-aws code tries to delete the "fake" directory test_run123/ it ends up deleting foo.bar. The real S3 SDK leaves the foo.bar file alone, and only tries to delete the "fake" directory, which should be expected behavior for this tool.