new data set: van Hateren Natural Images

[James Bergstra]

FYI you can now import skdata.vanhateren.dataset has some helper routines to download the data set from Paul Ivanov's mirror, check MD5s, and load it the images as numpy arrays. I would be open to adding routines for loading the patches, given some consensus on exactly how to do that. My sense is that local contrast normalization has mysterious constants, global contrast normalization seems potentially insufficient, multiple whitening strategies exist, etc., so there isn't anything completely obviously helpful that this file can do beyond loading the images into numpy.  Thoughts?

(Thanks Eric Hunsberger for writing the original code, which I adapted)

- James