ArrayTrain: BIC-selected unbounded θ mixtures + HWE-guided labelling

Fixed θ windows mislabelled compressed/asymmetric loci (2-cluster loci: 1418 vs
186 errors). Now 1/2/3-component mixtures chosen by BIC, labelled by HWE
goodness of fit plus weak position and gap priors, split clusters merged.
GSE165845 chr22 vs 1000G 30x (240 arrays): concordance 99.44 % (v1 99.27 %,
vendor EGT 99.46 %), NRD 1.30 % (1.89 %, 1.23 %), call rate 99.61 %.
docs/validation/V2_arraytrain.md, K-030; plan §12, CLI-038.

CHANGELOG.md

@@ -21,6 +21,11 @@ ClinPGx (70 genes); solver-free diplotype calling with array-equivalence classes
 DPWG phenotypes; per-sample JSON and the TSV set. The GSA v4 PGx demo arrays are 30 CEPH/1000G individuals: region
 CN 224/240 vs NYGC SV, gene calls 154/168 vs GeT-RM (`docs/validation/V2_pgx_cn.md`, `V2_pgx_star.md`, K-028, K-029).
 
+**ArrayTrain** (`simurg train`; plan §12): trains EGT cluster files (v3, dbv9) from a cohort's GTCs with sex-aware
+sample selection, BIC-selected 1/2/3-cluster θ mixtures, HWE-guided labelling, and missing clusters imputed from
+cross-locus relations. On GSE165845 (chr22 vs 1000G 30x) it gives 99.44 % concordance at a 99.61 % call rate;
+the vendor EGT gives 99.46 % at 99.15 % (`docs/validation/V2_arraytrain.md`, K-030).
+
 ## 1.1.0 — 2026-09-28 (P9, P10)
 
 **Packaging (P10).** `scripts/build_dist.sh` (sdist, pure wheel, manylinux_2_28 kernel wheels for x86_64 and