Krige: add switch to select variogram or covariance

[MuellerSeb]

Currently, the kriging matrix is build on the covariance function of the given model: https://github.com/GeoStat-Framework/GSTools/blob/21c97fc7426864add2430addcaf465da8993d227/gstools/krige/base.py#L290

To be future prove, we should add a switch to be able to select the variogram function as well.

For unbounded variograms (not supported at the moment) we can't express the kriging system by the covariance (not well defined then). Unbounded variograms are already supported by PyKrige and we need to support them: https://github.com/GeoStat-Framework/PyKrige/issues/136#issuecomment-605615590

Further, this could maybe be helpful for some numerical problems, but I am just guessing here.

[MuellerSeb]

Note

When setting unbiased=False (simple kriging), we can't use the variogram formulation, since the kriging weights need to sum up to 1 (unbiased condition) in order to transform the kriging system from the covariance formulation to the variogram formulation.

Consequently, simple kriging is not possible with unbounded models.