[ENH] refactor variogram fitting

[MuellerSeb]

The implementations for the variogram fitting got a bit out of hand: #364 (comment)

A solution could be a class for a variogram fitter. This could be combined with the variogram estimators to have a smoother workflow from sample data to a fitted model.

We could then also tackle these issues:

• [Enh] enable selcting axis during variogram fitting. #311
• [Enh] spatio-temporal variogram estimation #312