Automatic/standardized scale function computation

[SamTov]

What feature would you like to see added? Partially in conjunction with #383 but a different consequence, we need a standardized method of assessing the accuracy of the scaling functions.

My suggestion would be to set up a remote worker with ~4GB of memory, i.e. memory of a standard GPU, and check the scaling functions allow for correct operations on data that is both configuration-wise and atom-wise batched.