Support truncated singular value decomposition

[mdekstrand]

For many applications (recommender systems, information retrieval, etc.), only the top k singular values (and their corresponding vectors) are required from the singular value decomposition.

Would be nice to have the SVD methods support a decompose(AMatrix, int) method that specifies the (maximum) number of singular values desired, so we don't waste time and space computing a bunch of unneeded ones.

[mikera]

Sounds very sensible. I don't have time to implement this myself right now, but very happy to take PRs for this one.