Fix our Mersenne Twister implementation

[stiber]

Here is the original MT website, I think the most recent, from the Wayback Machine:

https://web.archive.org/web/20050316024259/http://www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/emt.html

Our issue is that this requires a separate program to be run ahead of time to generate the initial state for the desired number of MT threads. I believe that there is a replacement available from NVIDIA that may not require this.

Maybe this has already been fixed and not yet merged into SharedDevelopment? @AndrewBMadison ?

[stiber]

This is already solved by PR #859