Add team/serial SVD

[brian-kelley]

Implement a batched team-level and serial SVD.

This can be implemented using the algorithm described here:
http://www.netlib.org/utk/people/JackDongarra/etemplates/node193.html

We already have the other batched LAPACK functions (e.g. QR decomposition) needed by the algorithm.

[brian-kelley]

This feature's ready in #1107 . Just serial now, but everything's implemented in terms of existing batched kernels so it would be fairly easy to make a TeamVector version.