tskit input retains ancestral allele even when it has a count of zero

[molpopgen]

TODO: why are we keeping the ancestral allele

I think the test code expects the VCF reference allele to come first in the counts, yes.

But this is specific to VCF input, right? It is not a requirement of our site counts object in general?

Originally posted by @molpopgen in #100 (comment)

[molpopgen]

@laggycomputer -- converted this comment to an issue so that is isn't lost when #104 is merged.

[molpopgen]

I realized that we will need this for the case of polarized statistics (where it matters if an allele is ancestral or not). All of our stats so far are unpolarized.