Define an OpenML dataset loader

Currently, it is hard to publish papers using internal D3M datasets because one cannot really link to them. One way to address that is to allow also easy use of OpenML datasets directly. We could do that by defining an OpenML dataset loader.

The idea would be that one could provide a URI to OpenML dataset and loader would download it and load it. One could then use a Dataset saver to save it to D3M format for example.

We could extend Dataset.load interface with cache_path argument which if provided points to a directory where to save raw downloaded files and try to load from it instead of downloading from OpenML again. If cache_path is not provided, then OpenML dataset is downloaded again.

See !47 (merged) for initial attempt at this which could be continued.