biolink:encodes is a local coinage with no slot in the pinned biolink model

[realmarcin]

Surfaced in review of PR #341 (#334 burn-down). Pre-existing, not introduced there — 9
trait files use it — but the mapping row's note claimed something false, now corrected.

The gap

mappings/predicate_grounding.tsv grounds the label encodes to biolink:encodes, and
the note said "exact label match against biolink slot (RO:0002205 inverse: encoded_by)".

The pinned data/raw/biolink-model.yaml is 4.4.0 and has no slot named encodes — nor
encoded by. The gene→product slot is has gene product:

has gene product:
  domain: gene
  range: gene product mixin
  exact_mappings: [RO:0002205, ...]

So RO:0002205 belongs to a different slot, and biolink:encodes resolves to nothing
upstream.

The repo's own checker already records this: reports/biolink_coverage.tsv has the
encodes row as the only applied biolink: CURIE with both biolink_via_label and
biolink_via_exact_mapping empty. The signal existed; nothing acted on it.

Why it wasn't simply repointed in #341

A gene cluster → protein complex edge (mcrABG → methyl-coenzyme M reductase) is not a
clean fit for range: gene product mixin, which is presumably why the coinage exists in the
first place. Repointing would trade a missing term for a wrong range.

Options

1. Propose encodes to biolink (or adopt whatever they'd prefer for cluster→complex).
   Cleanest, but upstream and slow.
2. Mint a METPO predicate for it, in the v9 style — the corpus already has a place for
   causal-graph relations with no upstream home.
3. Repoint to biolink:has_gene_product/RO:0002205 and accept the range mismatch,
   or retype the nodes so gene→product holds.
4. Keep the coinage and document it, which is the current state after Re-ground 9 GENE_OR_PROTEIN-object enables edges by idiom (#334) #341 corrected the
   note to say plainly that it is local.

Worth checking the same way whether any other applied biolink: CURIE lacks a backing
slot — reports/biolink_coverage.tsv is the place to look, and it evidently is not being
read.

[realmarcin]

Additional scope from the #341 re-review, worth folding in here rather than doing piecemeal.

The machine-readable columns still say biolink. mappings/predicate_grounding.tsv's encodes row now carries a note saying plainly that it is a local coinage, but its source column is still biolink and its predicate_id column still skos:exactMatch. scripts/ground_causal_predicates.py reads source, so anything breaking grounding down by provenance will keep counting this as biolink-derived — the prose and the data disagree.

Fixing it properly means introducing a value like local to that column's vocabulary, which is currently METPO / biolink / RO / RDFS. That is a schema-ish decision affecting every consumer of the column, which is why it was left out of #341.

Also worth doing as part of this: sweep reports/biolink_coverage.tsv for any other applied biolink: CURIE with both biolink_via_label and biolink_via_exact_mapping empty. encodes was the only one at the time of writing, but nothing reads that report, so the same gap could reopen silently.