Improve usage of pykrige

[cchwala]

There are several issues with the usage of pykrige

• the version 1.3.1 which is the current one on pip and conda-forge, both seem to not correctly build the cython extension resulting in slow operation
• cloning the repository and using pip install -e correctly installs the cython extensions
• if using the windowd kriging option with nnear (which makes most sense for rainfall fields which can have very different semivariograms when interpolating over large area, e.g. like Germany) the results sometimes produce artifacts (possibly due to non-sense fits of the semivariogram) or pykrige even fails with a scipy.linalg error.

TODO:

• Force usage of pykrige > 1.4.0 (solved now, because pykrige has new version on conda)
• Catch errors if kriging fails or produces artifacts

[cchwala]

pykrige is now available in version 1.4.0 on conda-forge. This caused a problem, see #40. But the C backend should now work.

[cchwala]

I assume that we still do not catch the cases where kriging fails. Hence, I leave this issue open.