Random sampling

[redhog]

In cases where we have a very dense dataset, setting maxlag will not help the performance of the variogram calculation. Randomly sampling the points prior to creating the variogram is one option, but might lead to a high error in the variogram. It would be better in such a situation to randomly sample the point pairs.

Using a MetricSpacePair with two different random point samples instead of a MetricSpace would achieve this. It would need to be wrapped in something that makes the output of dists() square (and even more sparse).

What do you think of this solution?

[redhog]

Oh, this is a duplicate of #48