Handle recombinant parents given in unaliased form

[jchorl]

Problem

freyja demix --depthcutoff <n> crashes for any sample whose barcodes
collapse a group containing XHD:

File "freyja/utils.py", line 1125, in collapse_barcodes
    parent_aliases.append([lineage_data[lin]['alias'] for lin in parents])
KeyError: 'XDV.1.5.1.1.8.1'

collapse_barcodes looks up recombinant parents by lineage name, but
pango-designation's alias_key.json lists XHD's parents as
["XFG.3.4.1", "XDV.1.5.1.1.8.1"] — the second being the unaliased form of
NB.1.8.1, which is only ever an alias: value in lineages.yml, never a
name:. XHD is the only entry in alias_key.json with this inconsistency
(lineage_notes.txt correctly says "Recombinant lineage of XFG.3.4.1,
NB.1.8.1"), and it entered the barcodes with the 08_27_2026 update.

Low-coverage samples hit this readily: at --depthcutoff 1 with ~90% of the
genome at zero depth, nearly every barcode column is dropped and thousands
of lineages collapse into multi-class groups.

Changes

• Map recombinant parents back to lineage names when loading the hierarchy,
  so either form resolves.
• Extract per-group MRCA resolution into get_group_mrca(). A group that
  can't be resolved is now grouped into Misc with a warning instead of
  aborting. The previous except KeyError printed a message and fell
  through, leaving pango_aliases undefined (UnboundLocalError) or stale
  from the previous group, so a hierarchy behind the barcode file was never
  actually handled.
• get_path_to_root() promoted to module level (was redefined per iteration).
• Two regression tests using a small synthetic hierarchy, so they keep
  testing the code path if pango corrects the XHD entry upstream.

Testing

• New tests fail on main with KeyError / UnboundLocalError, pass here.
• flake8 clean; no change in test_utils results.

AI Disclaimer

I hit this issue and had AI help me resolve. I am fairly new to this codebase. If you are uncomfortable with this and don't have time to further investigate, feel free to close.