Add nodes argument to genotype_matrix

[hyanwong]

In #1202 (comment) @petrelharp suggested (and I agree) that it would be useful to add a 'nodes' parameter to the genotype_matrix() method, not only to restrict the size of the matrix returned, but also to be able to output ancestral genotypes/haplotypes (see #1206).

If this is done, I suggest that we might want to reimplement the haplotypes() method to use genotype_matrix() rather than iterating over the variants(), as this would probably remove some fiddly code and also automatically provide a way of getting ancestral (string-format) haplotypes, if we ever need that.

[jeromekelleher]

SGTM - should be a straightforward extension of the APIs in variants.

[hyanwong]

Great, thanks @jeromekelleher. I understand that you imagined it would be a bit fiddly to implement non-sample nodes together with isolated_as_missing=True in the variants code in C; I imagine this therefore also applies to the _ll_tree_sequence.get_genotype_matrix() method underlying the genotype_matrix() function. I'm just noting it in this issue for reference, as the ability to output ancestral haplotypes (with missing sections either side) was my original motivation for bringing this up.

[jeromekelleher]

I'm not sure there's actually much of a perf-gain from doing the genotype matrix stuff in C - we can probably use the variants() method in Python, which would simplify things.

[petrelharp]

I think this is a duplicate of #678?

[jeromekelleher]

Not quite @petrelharp - there we're talking about having a sites argument as well, which would require random access to the trees to really do properly.