use unphased xos to extract sites for phasing

[brentp]

given all the crossovers for unphased data, we can extract informative sites, phase only those, and the we can know the child in which the crossover was observed.

Informative sites should be the same more/less between methods.

This will give too many sites (will try to keep < 10M genome-wide) so we can filter to sites that are informative in >= N families based on the count in the .sites files that recombinator.py writes.

[brentp]

this is in progress in phased-from-unphased.sh though I'm hitting a lot of small problems that add up given the large dataset.