Species-overlap reconciliation mislabels duplications at gene-tree polytomies #117

Problem

LevelHOGProcessor.infer_reconciliation computes the species overlap of an internal node as the intersection of the species sets of all children (_infer_subhog.py:546). For a multifurcation (A1, A2, B1) the intersection over the three children is empty, so the node is labelled a speciation, although A appears in two children.

FastTree produces polytomies when sequences are identical or nearly so, i.e. exactly in recent duplications, so these get merged into one HOG where they should be paralogs.

Test (strict xfail on PR #112)

• tests/test_level_processor.py::test_reconciliation_polytomy_with_duplicated_species

Proposed fix

For nodes with more than two children, compute the overlap over pairs of children (duplication if any pair shares species; sos = max over pairs, or the union of the pairwise intersections divided by the union of all species), or resolve polytomies before reconciliation. This changes inferred HOGs, so regenerate the golden snapshots and review the diff.