All join count methods should warn/fail for asymmetric weights

[ljwolf]

Over in the geographic data science book, we just realized that join counts/local join count statistics are not well-defined in the case of asymmetric weights matrices (like k-nearest neighbors).

Thus, we need to warn/error out if w.asymmetries is not an empty list. I'm not sure how the statistics are defined in the directed case, and (regardless) we'd need to enhance their existing estimators for this case.

I think it's appropriate to raise a NotImplementedError immediately if the weights are asymmetric.

[sjsrey]

Is the issue the loss of the proportionality between the global JC and the sum of the local JCs with asymmetric weights, or something else?