Compose mixed inherited and somatic haplotypes (10.11.1)

[iskandr]

Mixed cis groups containing an allele already in GermlineContext previously returned Unresolved. They now compare novel edits against the patient baseline, retaining inherited alleles once. The #500 reproduction returns p.S159T rather than attempting to apply the inherited allele twice.

Normalize and deduplicate alleles once, then use the existing phase graph and layout pipeline for every edit type. Known-cis group members anchor phase before resolver answers; other germline alleles retain their uncertainty, and homozygous alleles do not bridge haplotypes. Results retain the original group membership, the first novel allele as the primary variant, and per-hypothesis phase/source/novel-edit evidence. All-inherited groups report allele overlap without inferring LOH.

Regression coverage includes inherited and novel SNVs/insertions/deletions, repeated and normalized-equivalent alleles, both input orders, reverse strand, cis/trans/unknown phase, phase propagation and limits, and the public collection/haplotype APIs. The scientific baseline follows Hundal et al.; direct haplotype inputs continue to assert cis, while collections require resolver support to form a group.

Validation:

• ./lint.sh passed.
• TEST_SH_MAX=4 ./test.sh: 3,152 passed, 34 warnings.
• mkdocs build --strict passed.
• Original Compose mixed inherited and somatic haplotypes without reapplying inherited alleles #500 reproduction: one Substitution p.S159T candidate, inherited allele cis.

Version: 10.11.1. Fixes #500. Published to PyPI from clean main; wheel/sdist SHA256 hashes match the local build, and an isolated install passes the original reproduction (p.S159T).

Release-process discrepancy encountered during inspection is already tracked in #414: the current deploy.sh does not implement the bump/guards/tags described in AGENTS.md. The version bump is included here; deployment ran from a verified clean main after CI and merge, following RELEASING.md.