[Enh] spatio-temporal variogram estimation

[MuellerSeb]

At the moment we provide directional variogram estimation, but we can't estimate an omni-directional variogram in a sub-space of the given data (eg. time-slices for a given spatio-temporal data set).

Implementing this could be quiet easy:

• We only need a tolerance for the slices in one direction (delta-t)

To not further bloat the vario_estimate routine, we could add a temporal_vario_estimate to estimate (omnidirectional) variogram of spatial dimension and separately the time dimension.

This could also handle the special case of spatio-temporal latlon data and is very useful for unstructured data.