Add better selection/acquisition functions

[ml-evs]

As discussed, @VicTrqt will play around with different acquisition functions "offline" -- i.e., use the existing dataset with fake values (noisy versions of the predictions) to explore the best approaches.

Tests can be added to make sure the functions fit the API.

One method we definitely want to try is the most-isolated Pareto solution (MIPS), from https://github.com/tsudalab/RPPF and https://journals.aps.org/prmaterials/abstract/10.1103/PhysRevMaterials.7.093804