Use spark's built-in SVD

[d-v-b]

Recently Spark added a built-in implementation of the SVD as a method on the RowMatrix class. I'm interested in running comparing the performance of this algorithm against the SVD implemented in thunder-factorization. If the Spark version is favorable, we should consider adding support for it in the factorization algorithms that use the SVD

[d-v-b]

ping @freeman-lab, @jwittenbach
My initial tests indicate that the pyspark SVD is much, much faster than the implementation in SVD.py. What would I need to show / do to get a PR reviewed that leverages the spark SVD for thunder-factorization?