The gene-symbol route is a second route on an adopted source and the queue cannot say so

[realmarcin]

Adversarial review of #260.

curation/source_queue.tsv records UniProt Subcellular Location as ADOPTED.
#260 adds scripts/uniprot_genes.py, which uses a different UniProtKB route
with different guarantees: it matches a gene symbol rather than a curated
localisation, and it carries no ECO:0000269 evidence for the structure — only
for whatever localisation the entry happens to state.

Nothing in the queue distinguishes them. A reader asking "is UniProt adopted?"
gets yes, and cannot tell that the two routes make different claims — the same
ADOPTED-cannot-express-coverage problem already recorded in #176.

Not a blocker for #260. Either the queue grows a row for the second route, or
docs/SOURCE_QUEUE.md says how a source with several routes is recorded.

[realmarcin]

Sharpening this after a second review pass on #260 — it is not only a
documentation question, there is a gate asymmetry behind it.

check_source_queue.py enforces, for every ADOPTED row:

if sid not in pipeline_sources:      # conf/sources.yaml must list it
if not row["script"]:                # the row must name a script
elif not (REPO_ROOT / row["script"]).is_file():   # that script must exist

Every check runs row → script. Nothing runs script → row. So a new script
that ingests from an adopted source is invisible to the queue, and just qc
passes.

scripts/uniprot_genes.py is exactly that: it pulls UniProtKB and writes
protein_examples, and it is named by neither curation/source_queue.tsv nor
conf/sources.yaml, both of which carry a single script: per source
(uniprot_sl → scripts/uniprot_sl.py).

The queue therefore says UniProt is adopted with one route and one script, while
the corpus has two routes making different claims — the SL route carries
ECO:0000269 evidence for the localisation, the gene route carries only gene
identity. Either the schema grows more than one script per source, or a second
row exists for the second route, or a gate checks that every ingesting script is
claimed by some row.