Fix source of randomness in doctests

[gdalle]

For instance with StableRNGs

[simonschoelly]

One issue I see with StableRNG is that we then might have to put that also in the example code. Then users might think that they should always use StableRNG - but if there is a way to somehow hide that, then yes, we can use it.

[simonschoelly]

Another - ugly method - would be to have different docstrings for different Julia versions.

[gdalle]

Or to filter docstrings with regexes but that's a whole nother can of worms that I don't want to open. I think StableRNGs is the least worst option, but I agree it's less than ideal.

Doctests are very irregular in the code base anyway, and I don't think that many methods have them. Would it make sense to remove them and add tests instead?