MersenneTwister: implement generation of jump-ahead polynomials

[rfourquet]

This completes PR #12498, which provided the jump-ahead functionality, by allowing to specify an arbitrary number of steps instead of the hard-coded 10^20. This code may very well be better-suited for a package, but it's so tiny (roughly the same size as the original code from #12498) that I thought I could push it here as well.

This is WIP as I'm considering evolving the randjump API, e.g. deprecating the method randjump(rng::MersenneTwister, jumps, jumppoly::AbstractString). Also I think I would like to have randjump(rng, steps) (and randjump!(rng, steps)) but this is incompatible with randjump(rng, jumps).

[tkelman]

need to double check this, but having the comment between the docstring and the const might mean the docstring is not associated with the const?

[rfourquet]

I just checked and it works! but I'll move anyway the comment after the definition.

[wildart]

You can make jumps parameter in randjump to default to 1. And add steps as keyword argument so

randjump(rng, 10) # ten jumps with default steps
randjump(rng, steps=10) # one jump with 10 steps
randjump(rng, 10, steps=10) # ten jumps with 10 steps each

It's the same thing for randjump!.

[tkelman]

maybe explain in words what GF(2)[X] means?

[rfourquet]

Updated.

[rfourquet]

I think I agree with your proposed API for randjum. I will remove the WIP, as I now think it's better to discuss an updated API in a new PR, which will depend on wheter this present PR gets merged or not.

[rfourquet]

This is freshly rebased. Barring objections, I will just merge this: this adds only a little bit of code, and adds the missing half of the randjump functionality. And with Random moving to "stdlib", it makes even more sense, as opposed to putthing this into a separate package. Moreover, this will be useful as a complement of #25129.