~11 `reduces` edges use the 'lessens' sense but sit on the redox term METPO:2007802

[realmarcin]

Raised in review of PR #329 and split out at the reviewer's suggestion rather than
growing that PR further. Not fixed there — but deliberately not silent either: the
reduces row in mappings/predicate_grounding.tsv now documents both senses and links
here, so it does not land as unexamined precedent.

The problem

METPO:2007802 (reduces, proposed in proposals/metpo_traitmech_v9) is defined as:

asserts that the subject donates electrons to the object, lowering its oxidation state

30 corpus edges carry it, and they are two different senses:

sense	count	example
redox (correct)	~19	terminal_oxidases --reduces--> molecular_oxygen
"lessens / decreases"	~11	membrane_lipid_remodeling --reduces--> inward_proton_leakage

The second group asserts something false under the term's definition — membrane lipid
remodelling does not donate electrons to proton leakage. Full list of the "lessens" group:

• ecology/gut_associated.yaml — intestinal peristalsis → colonization/retention in duodenum
• environment/acidotolerant.yaml — membrane lipid remodeling → inward proton leakage
• environment/facultatively_acidophilic.yaml — hopanoid biosynthesis → membrane proton permeability
• environment/nacl_range.yaml — cyclic di-AMP → high-salt tolerance
• environment/ph_delta_high.yaml — membrane lipid/porin composition changes → inward proton leakage
• environment/ph_delta_mid3.yaml — membrane lipid composition shifts → membrane proton permeability
• environment/ph_range_very_low.yaml — inside-positive membrane potential → passive proton influx
• environment/temperature_optimum_high.yaml — positive DNA supercoiling → DNA melting at high temperature
• genomics/plasmid_carriage.yaml — compensatory mutation → plasmid fitness cost
• morphology/gram_variable.yaml — thin peptidoglycan layer → CV-iodide complex retention
• physiology/oligotrophic.yaml — streamlined genome → reduced metabolic redundancy

Note oxygenic_photosynthesis.yaml's photosystem I --reduces--> ferredoxin is genuinely
redox despite its non-chemical object, so "object is not a CHEMICAL" is not a safe
classifier — this needs reading per edge.

Why (predicate_id, label) cannot separate them

The migration script keys on that pair precisely because METPO:2000013 is reached from
two labels with opposite directions (#327). Here the label is identical — both senses
are written reduces — so no key over the existing data distinguishes them. Fixing it
requires a corpus relabel first.

Suggested fix

1. Relabel the ~11 "lessens" edges to a distinct predicate grounded to RO:0002212
   (negatively regulates), which already has a row in mappings/predicate_grounding.tsv.
   Each needs reading — several may be better served by RO:0002212 vs a "decreases
   quantity of" sense.
2. Narrow the reduces row's subject_types/object_types to the redox shape once the
   others have moved.
3. proposals/metpo_traitmech_v9/metpo_proposal_properties_robot.tsv — the METPO:2007802
definition_source cited the non-redox peristalsis edge as its exemplar. Fixed in
   Migrate the remaining microbe-domain edges (#301, part 2 of 2) #329 so the upstream artifact does not carry a wrong exemplar, but noted here for
   completeness.

Related: #327 (the other predicate whose corpus usage does not match its term's sense).