Enable path based versioning of datasets in cz-benchmarks

[mlgill]

Currently, versioning of dataset files in cz-benchmarks must be done by changing the dataset name. This is because the dataset file is downloaded directly into CACHE_DIR/datasets and the download_file_from_remote function uses path to determine if the dataset file needs to be downloaded. One option to remedy this would be to preserve at least the part of the URL with the version number in it when mapping to a location in the cache.