Explore if preliz can benefit from probabilistic numerics routines

[aloctavodia]

Maybe some of our functions could benefit from a Probabilistic Numerics perspective. https://www.probabilistic-numerics.org/

Functions like maxent or roulette could either return a single distribution (current behavior) or a few distributions representing the optimization uncertainty and input uncertainty. For example, for the roulette method, the discreteness of the bins/chips.

The main goal is to avoid users to be overconfident of their decisions.