Optimize GHC for larger datasets

[juliohm]

The current implementation uses sparse matrices to handle large datasets with low range parameter, but preliminary steps still use dense matrices. We need to optimize the memory usage and then decide on a maximum number of samples above which we create a pipeline with sub-sampling and nearest-neighbor interpolation.

[juliohm]

Fixed in GeoStatsTransforms.jl v0.10.2 with the introduction of a new nmax option. The option can be used to sub-sample the input geotable and avoid unbounded memory consumption. The result is then interpolated with nearest neighbors to the original domain.