Consider allowing user to specify RNG seed in config

[drewoldag]

I'm really on the fence about this particular thought.

Pros:
Reproducibility

Cons:
Easy for someone to forget/copy and introduce a subtle "bug" that could be very hard to figure out

For something like demonstrations or testing, this could be very helpful, but again, it leaves the door open for mistakes that would otherwise be easy to avoid.

[mtauraso]

Doesn't completely address the con, but we could Have two configs: use_system_rng (t/f). and rng_seed (number). We could preference rng_seed_from_system over rng_seed.

The idea here is that in order to get the "dangerous to science" behavior of having a set rng seed, you have to change 2 values, but in order to verify that you have the "safe for science" behavior of using the system prng, you only have to check 1 value (use_system_rng).

The defaults would be: use_system_rng: true, rng_seed: false where "false" falls back to the system RNG unless the user specifies a seed.

In cases where the user specifies only one value there would be an error message designed to get them on the right track.