[WIP] Change RNG seed generation scheme.

[ye-luo]

step 1, get initial seed by time std::chrono
step 2, generate seeds for RNGs used by MC. Each rank needs omp_get_max_threads()+1 seeds. jump ahead rank()*(omp_get_max_threads()+1) with discard(). Very fast 0.14 sec for 10^8.
step 3, seed the per-rank RNG.
step 4, seed the per-thread RNG.

After dive in the code a bit, I feel a lot of cleaning is needed.
Mostly RandomGenerator_t needs to be simplified. All the information about MPI and threads should be removed.

[prckent]

1. Agree that lots of cleaning is needed.
2. I would prefer that we don't touch this shortly before a release and only touch it once. There are many infamous incidents of initialization being done incorrectly.
3. We should move to a scheme that guarantees independent random numbers i.e. different seeds per thread. There is a formally correct way to do this that I outlined earlier today.

[prckent]

See #1176

[jtkrogel]

What is the typical resolution of "high_resolution_clock" in std::chrono?

Would really like the probability of repeated clock based seeds to be near zero (e.g. no repeats as in #1068, but guaranteed for an even larger set of runs).

[PDoakORNL]

why not use a hardware random source? I think openssl offers easy access to hardware random bytes on any platform we use? Take the seed from there, you can still record it but you shouldn't have to worry about collisions.

[prckent]

I have some better ideas on how to handle this now. A good topic for v4 but we need the batched code first.