Add span_normalise to divergence_matrix

[jeromekelleher]

As discussed by @petrelharp here: #2813 (comment)

Also, I think we should add the span_normalise argument (and probably have it default to True, as usual). (Here's how we do it elsewhere.) The reason is that, if I've got this right, currently the code is computing the non-span-normalized version (ie just counting mutations);l however, "divergence" usually means "number of mutations divided by sequence length", i.e., span normalized. But we'll have need for the "just count mutations" version, so rather than do error-prone post-hoc multiplying by window lengths to get it out, we should just add the argument.