Feature: Include non-uniform distributions for `get_evaluation`

[lfrommelt]

Most real world data is not distributed uniformly. While often normally distributed data is a best case scenario for ML, normal distributions are better suited for evaluating real-world performance of benchmarked SR algorithms. It would be nice, if the user can give their own distribution as an argument to get_evaluation (caveat: random states), or if at least the normal distribution and potentially others are hard-coded and accessable by a string kwarg.