feat: add GTDB as a biolink:OrganismTaxon source in the taxon pipeline

[SkyeAv]

Adds GTDB (Genome Taxonomy Database, bacterial/archaeal taxonomy) as a
biolink:OrganismTaxon source in the existing taxon pipeline. Every rank in each genome's
gtdb_taxonomy string becomes a labeled GTDB CURIE (GTDB:s__Escherichia_coli), and species-rank
taxa are linked eq to their ncbi_species_taxid so a GTDB species resolves into the NCBI taxon clique.

Ingest

• New handler: src/datahandlers/gtdb.py parses the bac120/ar53 metadata (columns read by
  name, robust to reordering). write_gtdb_labels emits one label per unique taxon (all ranks);
  build_gtdb_relationships emits species-rank GTDB↔NCBITaxon edges and records the configured
  download URL in the concord provenance.
• Download: get_gtdb_bac120 / get_gtdb_ar53 rules (retries: 3) via pull_via_wget;
  gtdb_metadata_url_prefix in config.yaml defaults to GTDB releases/latest/ and should be pinned
  for a reproducible build.
• Wiring: gtdb_labels / gtdb_ids / get_gtdb_relationships in taxon.snakefile; GTDB added to
  taxon_labels / taxon_ids / taxon_concords. GTDB joins the existing OrganismTaxon.txt (no new
  compendium file), so the util.py aggregators and report_tables.py need no change.

Design (load-bearing)

• extra_prefixes=[GTDB]: GTDB is not in the Biolink Model's organism taxon id_prefixes
  ([NCBITaxon, MESH, UMLS], verified against the pinned 4.4.3), so write_compendium would silently drop
  every GTDB CURIE; the taxon build passes extra_prefixes=[GTDB] (the documented escape hatch). Filing
  GTDB with the Biolink team is the long-term fix.
• GTDB in unique_prefixes: the GTDB↔NCBI species mapping is many-to-one, so two GTDB species sharing
  one ncbi_species_taxid must not merge. Listing GTDB as unique makes glom reject that merge — at most
  one GTDB species joins a given NCBITaxon clique (first-in-file-order), the rest stay singletons.
• Prefix casing: Babel uses display-cased GTDB; Bioregistry's preferred prefix is lowercase gtdb.
  GTDB is not yet in the Biolink Model, so there is no conflict — reconcile if it is registered upstream.

Docs

• docs/sources/GTDB/README.md (new) + an entry in the docs/sources/ index.

Testing

• uv run pytest -m unit tests/datahandlers/test_gtdb.py -q --no-cov → 8 passed — over a verbatim
  113-column GTDB fixture: taxonomy parsing (spaces→underscores, label spacing), one-label-per-taxon + dedup,
  species-only GTDB↔NCBITaxon concord, by-name column lookup, gzipped input, missing-value skips, and a
  glom-level test pinning the many-to-one guard (two GTDB species sharing an NCBI taxid stay two cliques).
• uv run pytest -m network --network tests/datahandlers/test_gtdb.py -q --no-cov → 1 passed — create_node
  keeps a GTDB CURIE with extra_prefixes=[GTDB] and drops it without (pins the extra_prefixes decision).
• uv run pytest -m unit -q (full CI gate) → 410 passed.
• uv run snakemake -c1 taxon -n --quiet → DAG builds (all 5 GTDB rules wired); ruff / snakefmt / rumdl
  → clean.

Caveats / Deferred

• License: GTDB is CC-BY-SA-4.0 (share-alike) — confirm acceptable for downstream redistribution.
• Coverage: species with no NCBI equivalent and all higher ranks form labeled singletons.
• Comparison deferred (per maintainer): no source-impact report / clique diff yet. The taxon pipeline has
  no impact-report hook (PIPELINE_CONFIG registers only anatomy/disease); registering one + a build-vs-build
  clique diff is the recommended follow-up, since the many-to-one design is exactly what that scrutiny is for.

Questions for the reviewer

• Many-to-one handling. Is "at most one GTDB species per NCBI taxid (first-in-file-order), rest singletons"
  the right tradeoff, or should species-rank GTDB↔NCBITaxon links be omitted until a 1:1 mapping is available?

---

AI-drafted: This PR was drafted by an AI coding agent (pi) at the maintainer's direction and is pending
human review. Verification commands above were run by the agent.