chrX marker recovery for sex-matched host/donor pairs

[davmlaw]

The panel carries ~27 chrX amplicons that we default-drop. For a host/donor pair of the same sex they are usable informative markers; inferring sex from the data and keeping chrX when it matches adds markers, which matters most at low fractions where every informative marker counts.

Evidence: [speculative]. Low priority. Was part of Step 34 in the retired overall plan. (The other Step 34 item, input-quality QC predicting poor fits, is covered by the per-sample coverage-uniformity + allele-balance checks in #38.)

[davmlaw]

Scoping note: this issue covers the sex-matched case only, where it is genuinely easy because chrX carries the same ploidy in host and donor, so those markers behave like autosomes and drop straight into the existing diploid model. No estimator change needed. The exploit-the-mismatch idea is a different mechanism and is tracked separately (see #48).

What "let chrX through when it's safe" needs, in order:

1. Infer sex from the data for host and each donor. No sex inference exists today (only a free-text --recipient-sex metadata field, shown verbatim). Infer from the genotypes we already load: chrX heterozygosity rate (males ~0 het on non-PAR chrX) and/or chrY marker depth relative to autosomes if the panel carries any chrY amplicons.
2. Confirm the pair matches, and only then keep chrX. If host and donor sex differ (or either is ambiguous), keep dropping chrX. The whole benefit rests on matched ploidy; a mismatch makes the /2 dosage wrong (already noted in the --use-sex-chroms help).
3. Exclude PAR1/PAR2. Pseudoautosomal regions are diploid in both sexes, so a chrX/chrY amplicon inside PAR would be miscounted under a hemizygous assumption. Mask by coordinate (build-specific PAR intervals) before letting anything through. For the both-male-matched case chrX is hemizygous, so either restrict the easy win to both-female pairs (chrX diploid in both, cleanest) or handle male-male chrX as ploidy-1-in-both.
4. Guard bias estimation. Per-marker bias is estimated at het sites assuming expected VAF 0.5. Any spurious het on hemizygous male chrX would corrupt that, so exclude non-PAR chrX from bias fitting when the sample is male.
5. Keep the existing "informative sex-chrom markers dropped" reporting so the recovered count is visible.

Suggested smallest first cut: infer sex, and for same-sex pairs with chrX diploid in both (i.e. both female, PAR excluded), route chrX through the normal path. That captures most of the ~27 chrX amplicons' value with no change to the dosage math. Male-male hemizygous chrX and any mismatch handling can come later / live in #48.