Submodule for datasets

[mauro-milella]

This pr addresses issue #40.

This is what I have done until now:

• define 4 loaders for different datasets (HuGaDB, NATOPS, Libras, Epilepsy);
• each loader is enriched with a docstring, explaining where to download the datasets and, of course, the meaning of each Arguments and Keyword Arguments;
• define a submodule, Datasets, which exports the loaders;
• include the submodule in SoleData, and @reexport .Dataset;
• move an old "load_arff_dataset" dispatch from SoleData to within .Datasets submodule: this dispatch is still widely used in tests, but essentially, it is a messy version of load_NATOPS loader.

[mauro-milella]

Some tests are not working, but #39 should fix them.

To test the loaders I propose, it would be necessary to load them inside test/data. Can we do this?
This could be an overhead too big for certain datasets (e.g., HuGaDB is 220MB).
Maybe we could crop them to just a few instances?

[alberto-paparella]

Dear @mauro-milella, yes using only a smaller set (warning the user) of the dataset could be a good idea.