GPU acceleration by parallel diagonalization

[anjohan]

This branch includes the batched diagonalization in the phonon and electronWannier Hamiltonians, which are called from the bandstructure constructor.

Note: Most of the eigenvector unit tests have been disabled. Everything "works" in the sense that the thermal conductivity comes out correct, and there is freedom of both phase and degenerate subspace linear combinations, making the eigenvectors hard to test.

Note: Only properly tested on our limited test cases (silicon etc.).

Future enhancements that may be necessary:

• The same parallelization in buildAsPostProcessing
• The same parallelization in buildOnTheFly
• More careful batching in case of memory issues/limitation
• More comments

[jcoulter12]

Ok, other than the few places I noted on the PR that it would be good to have another comment, and the note that we should check GaN band structure since the long range correction was edited, this looks good to me!

Thanks @cepellotti and @anjohan for all the work on this PR!

[anjohan]

I've added some more comments.

The electronWannierTransport has also been re-checked for GaN and shows good agreement with the develop branch.