Implement some more data loading utilities

Real data:

* http://scikit-learn.org/stable/modules/generated/sklearn.datasets.fetch_mldata.html#sklearn.datasets.fetch_mldata
* http://scikit-learn.org/stable/modules/generated/sklearn.datasets.fetch_rcv1.html#sklearn.datasets.fetch_rcv1

Synthetic data:
* http://scikit-learn.org/stable/modules/generated/sklearn.datasets.make_low_rank_matrix.html#sklearn.datasets.make_low_rank_matrix
* http://scikit-learn.org/stable/modules/generated/sklearn.datasets.make_sparse_spd_matrix.html#sklearn.datasets.make_sparse_spd_matrix

This would be useful: too often I make my test matrices with `da.random.normal`, which is far from a realistic test. These are a little better, and would be a lot easier to implement.

I don't have any pressing need for real datasets I've listed, but the http://mldata.org one is very general and the RCV1 could be useful (many features, many examples, sparse dataset).