Tracking: bugs found by the test suite ahead of the ete4 migration #124

PR #112 adds a test suite (unit, stage and nf-tests, plus golden snapshots on 23 rootHOGs of the fungi-30 dataset) in preparation for migrating from ete3 to ete4. Known bugs are pinned as xfail(strict=True) tests with a file:line reason; pytest -ra lists them. Each issue below names its tests. An issue is done when those tests pass and their xfail markers are removed (strict xfails fail CI as soon as they pass, so the markers can't be forgotten).

Bugs, in suggested order

• infer-subhogs silently drops proteins (3% of all proteins on fungi-30) #113 infer-subhogs silently drops proteins (3% of all proteins on fungi-30)
• infer-roothogs silently drops proteins #114 infer-roothogs silently drops proteins
• ImpliedLosses score is wrong: species tree nodes are compared by identity #115 ImpliedLosses score is wrong: species tree nodes are compared by identity
• Output orthoXML violates the OrthoXML 0.5 schema #116 Output orthoXML violates the OrthoXML 0.5 schema
• Species-overlap reconciliation mislabels duplications at gene-tree polytomies #117 Species-overlap reconciliation mislabels duplications at gene-tree polytomies
• Input validation: crashes and misleading errors in check-input and infer-roothogs #118 Input validation: crashes and misleading errors in check-input and infer-roothogs
• Output depends on PYTHONHASHSEED and on Nextflow task completion order #119 Output depends on PYTHONHASHSEED and on Nextflow task completion order
• Bring fragment handling back to life (--overlap-fragments has no effect) #121 Bring fragment handling back to life (--overlap-fragments has no effect)
• Small internal bugs in HOG class, UnionFind, batching and orthoXML helpers #122 Small internal bugs in HOG class, UnionFind, batching and orthoXML helpers
• Report fails when CompletenessScore is disabled #123 Report fails when CompletenessScore is disabled

Discussion

• Discussion: future of the gene tree rooting methods (mad, Nevers_rooting, outlier) #120 Future of the gene tree rooting methods (mad, Nevers_rooting, outlier)

How to work on these

• One PR per issue (Input validation: crashes and misleading errors in check-input and infer-roothogs #118 and Small internal bugs in HOG class, UnionFind, batching and orthoXML helpers #122 can each be one PR with several small fixes). Several fixes change the inferred HOGs or scores (infer-subhogs silently drops proteins (3% of all proteins on fungi-30) #113, ImpliedLosses score is wrong: species tree nodes are compared by identity #115, Output orthoXML violates the OrthoXML 0.5 schema #116, Species-overlap reconciliation mislabels duplications at gene-tree polytomies #117, Bring fragment handling back to life (--overlap-fragments has no effect) #121); keeping them separate keeps every golden snapshot diff attributable to one cause.
• Golden snapshots (tests/data/fungi30/golden/) record current behaviour, bugs included. After an intended change run python tests/fungi30.py regenerate (needs docker) and review git diff tests/data/fungi30/golden as part of the PR. They only compare in the snapshot's tool environment (mafft 7.505 / FastTree 2.1.11, as in the dessimozlab/fastoma image and Ubuntu 24.04 CI).
• ete4 migration: should run as a pure refactor against unchanged snapshots. Either fix the output-changing bugs first, or migrate first and fix afterwards, but don't mix the two in one PR. ImpliedLosses score is wrong: species tree nodes are compared by identity #115 (store taxon names instead of TreeNodes in HOGs) and Discussion: future of the gene tree rooting methods (mad, Nevers_rooting, outlier) #120 (removing unused rooting code) also shrink the ete3 surface that has to be migrated.