switch SLiM's random number generator (RNG) to be faster and better

[bhaller]

Right now SLiM is using the taus2 RNG included in the GSL for 32-bit draws, and a Mersenne Twister variant for 64-bit draws. These generators are quite old, slow, and (at least in the case of the MT generator, not sure about taus2) statistically imperfect. The PCG family of generators at https://www.pcg-random.org/index.html is much faster (about 3x, by my testing) and statistically very strong, and provide some other benefits that might prove useful as well. I should switch SLiM over. For simulations that lean very heavily on random number generation, this might produce a user-visible performance improvement.

Note that PCG lives on GitHub at https://github.com/imneme/pcg-cpp and that repo has a more recent version of the code than what is provided by the downloads page at https://www.pcg-random.org/index.html. I will use the repo version of the code. There are numerous pull requests on the repo that have not been integrated; I will not adopt any of them unless I hit compile problems somewhere.

[bhaller]

Perhaps worth noting: I've been thinking about switching RNGs for a long time, but I thought I was limited to the RNGs provided by the GSL, of which the taus2 generator that I have been using seems the best. Then, just now, I realized that I can actually wrap my own generator in the same way as the built-in GSL generators, and then the GSL will happily use it. So now the PCG's pcg64_fast generator is wrapped and used by the GSL, giving us one shared generator for both GSL and non-GSL use, which is great. I also use a pcg32_fast generator, seeded identically, for places where a 32-bit random draw suffices (since it is ~10% faster than pcg64_fast).