Remove inference and subset-comparison arrows from Parmigiano set2 graph #1678

CommunityMech:000431; PMID41859451. Preserve four biological assay nodes, positive St/Ld pair versus both monocultures, lower Lh/Ld pair rates, highest triculture total lactate and set2 St/Ld rate comparable to triculture. Remove acidification-to-inferred-C5I3-activity arrow and pair-to-triculture comparison arrow: neither is a demonstrated causal process. Rename D-lactate node to the measured triculture pool, not a specific producer; Lh and St can also contribute D-lactate. Preserve #1077-1079 fixes and existing molecular mechanism gap. Close only after protected-main merge.

Searched 954 all-state issue titles/bodies; read all five matches (#1077-1079, #497, #258), the previous milk review, and the existing Polaromonas author-request lead. Local hidden/ignored-inclusive search found 64 matching files, excluding .git/docs/site/references_cache. Four whole records and four fresh complete primary main texts/captions read; seven further abstracts checked. No original figure pixels or separate supplements audited. Self-adversarial review, not independent approval.