Remove Neo4j ID set filtering from curation indexers

[oblodgett]

Summary

• Remove legacy Neo4j ID cross-reference filtering from all curation-sourced indexers
• The curation API is now the authoritative data source; the Neo4j ID intersection was filtering out valid curation entities that hadn't been loaded into Neo4j
• Existing obsolete/internal entity checks remain as the data quality filter
• BaseService no longer has any Neo4j dependencies (removed AlleleRepository, GeneRepository, VariantRepository usage)

Indexers updated

• GeneToGeneOrthologyIndexer — removed objectGene Neo4j filter
• GeneMolecularInteractionService — removed interacting genes Neo4j filter
• GeneGeneticInteractionService — removed genes + perturbating alleles Neo4j filter
• GenePhenotypeAnnotationService — removed subject gene Neo4j filter
• AllelePhenotypeAnnotationService — removed subject allele Neo4j filter
• AGMPhenotypeAnnotationService — removed subject AGM Neo4j filter
• SiteMapAccessionCurationIndexer — removed retainAll() intersection with Neo4j gene/allele ID sets

Test plan

• Deploy to stage and run a full indexer pass
• Verify indexed document counts are >= previous counts (more entities expected since Neo4j filter no longer drops valid curation entities)
• Spot-check orthology, interaction, phenotype, and sitemap results in the stage UI

[claude]

Code Review

The changes look correct. This is a clean removal of the Neo4j ID cross-reference filtering from all curation indexers, making the curation API the sole authority for entity validity.

Verified:

• No remaining callers of the removed methods (getAllNeoGeneIDs, getAllNeoAlleleIDs, getAllNeoModelIDs, getAllNeoVariantIDs, isValidNeoEntity, hasInteractingGenesInNeo, hasPerturbatingAllelesInNeo) anywhere in the codebase.
• BaseService still has subclasses (BaseInteractionService, BaseDiseaseAnnotationService) that use its remaining functionality (readFromCache, writeToCache, hasNoExcludedEntities), so it's not dead code.
• The obsolete/internal entity checks are correctly preserved as the remaining data quality filter.
• No new logic introduced — purely subtractive change, which limits the risk surface.

No issues found.