fix(lib): derive GT phase from allele strings vs ref column

[Jaureguy760]

Summary

The GT parser previously only recognized VCF-standard "0|1" and "1|0" formats. Input data with allele-string GT values (e.g., "A|C", "G|T") was silently treated as unphased, causing per_donor phased analysis to produce identical results to unphased.

Changes

Modified rust/src/lib.rs to support two GT formats:

1. Numeric (VCF standard): "0|1" or "1|0"
2. Allele strings: "X|Y" - derives phase by comparing to the ref column
   • If first allele == ref → gt = 0 (ref on hap1)
   • If second allele == ref → gt = 1 (ref on hap2)

Verification

Version	per_donor phased λ_GC	per_donor unphased λ_GC	Diff
OLD	1.501	1.501	0.000 (BUG)
FIXED	1.106	1.501	0.395 (FIX WORKS)

• OLD: phased == unphased (λ_GC diff = 0.000) ❌
• FIXED: phased ≠ unphased (λ_GC diff = 0.395) ✅
• Phased λ_GC (1.106) closer to 1.0 (well-calibrated) than unphased (1.501)

Test plan

• Quick test on 1000 rows: 35 regions with different p-values (vs 0 for OLD)
• Full parity grid on 1.26M rows, 75k regions
• λ_GC patterns statistically sensible (phased better calibrated)
• Existing unit tests pass (cargo test)

🤖 Generated with Claude Code