[v0.2] Multi-witness master alignment graph (Epic)

[bsesic]

Stage 2 of the long-term TRACE vision (see `docs/ROADMAP.md`): simultaneous multi-witness alignment producing a canonical variant graph plus a derived aligned table view.

Goal

Implement `tracealign.align_multi(witnesses, lang, config) -> MultiAlignmentResult`, building on the v0.1 pairwise aligner. Three-phase pipeline:

1. Pairwise distance matrix via v0.1 `align()`.
2. UPGMA guide tree built from the distance matrix; deterministic tie-breaking on sorted witness IDs.
3. POA-based progressive merge: walk the guide tree in post-order, align each new witness against the current variant graph via partial-order alignment, merge the result.

Phase issues

• [v0.2] Phase 1: Bootstrap and data model #2 — Phase 1: Bootstrap and data model
• [v0.2] Phase 2: Pairwise distance matrix #3 — Phase 2: Pairwise distance matrix
• [v0.2] Phase 3: UPGMA guide tree #4 — Phase 3: UPGMA guide tree
• [v0.2] Phase 4: POA-based progressive merge #5 — Phase 4: POA-based progressive merge
• [v0.2] Phase 5: Public API #6 — Phase 5: Public API
• [v0.2] Phase 6: JSON persistence #7 — Phase 6: JSON persistence
• [v0.2] Phase 7: Property tests, E2E, documentation #8 — Phase 7: Property tests, E2E, documentation

Reference documents

• Design spec: `docs/superpowers/specs/2026-05-21-trace-v0.2-multi-witness-design.md`
• Implementation plan: `docs/superpowers/plans/2026-05-22-trace-v0.2-multi-witness.md`

Branch and release strategy

• Feature branch: `feature/v0.2-multi-witness` off `develop`
• PR target: `develop`
• Release branch for cutting 0.2.0: `release/0.2.0` off `main`

Acceptance criteria (from spec §8)

• `align_multi(witnesses, lang="hbo", config=None)` works for any well-formed input
• `flake8 src/ tests/` is clean
• Full test suite passes on Python 3.10 / 3.11 / 3.12 (CI matrix green)
• Permutation-invariance property test passes
• Lossless-reconstruction property test passes
• JSON round-trip via `tracealign.io.multi_result` preserves all fields including the guide tree's distance matrix
• Documentation updated (`usage.md`, `details.md`, `ROADMAP.md`)
• 0.2.0 tagged on `main`, uploaded to PyPI, archived on Zenodo
• `CITATION.cff` version bumped

[bsesic]

v0.2.0 released and live. All acceptance criteria from spec §8 met:

• ✅ align_multi() works for any well-formed input
• ✅ flake8 src/ tests/ clean
• ✅ pytest green on Python 3.10 / 3.11 / 3.12 (182 tests)
• ✅ Permutation invariance property passes
• ✅ Lossless reconstruction property passes
• ✅ JSON round-trip via tracealign.io.multi_result preserves all fields including the guide tree's distance matrix
• ✅ Documentation: docs/usage.md, docs/details.md, docs/ROADMAP.md (Stage 2 marked released), docs/index.md, docs/faq.md, README.md
• ✅ 0.2.0 tagged on main, uploaded to PyPI (https://pypi.org/project/tracealign/0.2.0/)
• ✅ Zenodo archive: https://doi.org/10.5281/zenodo.20440062 (with ORCID linked)
• ✅ CITATION.cff version + DOI bumped on main and develop

Develop is now bumped to 0.3.0.dev0 for the next stage. Closes #9.