Fix source relationship direction and duplicate inverses

[ewlarson]

Summary

• label a child's dct:source link as Source record...
• label a parent's inverse source links as Derived records...
• standardize incremental relationship sync on the existing canonical dct:isSourceOf predicate
• normalize and deduplicate legacy dct:sourceOf rows while they remain in the database
• make Browse all for derived records filter children through dct_source_sm

Before and after

The original issue identifies this parent and child:

• Parent: Urban Base Layers [Maryland--Baltimore] {2026}
• Child: Address points [Maryland--Baltimore] [2026]

Parent resource

Production currently returns both dct:isSourceOf and the legacy dct:sourceOf for the same 20 children.

Before	After this change
Source records... → Address points, Bridges, Bus routes, etc.	Derived records... → Address points, Bridges, Bus routes, etc.
A second dct:sourceOf section repeats the same children.	The legacy alias is normalized and duplicate children are removed, leaving one section.
Browse all 20 records... falls back to a nonexistent dct:isSourceOf_agg filter.	Browse all 20 records... uses include_filters[dct_source_sm][]=b1g_gxSjhzRCQ7xQ.

Child resource

The child stores b1g_gxSjhzRCQ7xQ in dct_source_sm, which means the linked record is its source/parent.

Before	After this change
Derived records... → Urban Base Layers [Maryland--Baltimore] {2026}	Source record... → Urban Base Layers [Maryland--Baltimore] {2026}

The Princeton example from the issue currently has the same duplicate inverse predicates; it is normalized by the same compatibility path.

Root cause

The frontend labels for dct:source and dct:isSourceOf were reversed. Separately, the full relationship population script used dct:isSourceOf, while newer incremental synchronization emitted dct:sourceOf. A full reconciliation therefore left the old canonical row in place and added a second alias row.

Incremental synchronization now emits dct:isSourceOf and includes dct:sourceOf in its cleanup set. The relationship presenter also canonicalizes and deduplicates both spellings, so deploying and refreshing resource representations corrects existing records without requiring an immediate full relationship rebuild.

Validation

• python -m pytest backend/tests/services/test_relationship_sync.py backend/tests/services/test_relationship_service.py — 21 passed
• make lint-check — passed
• npm test -- --run — 946 passed
• npm run build — client and SSR production builds passed
• targeted frontend ESLint and Prettier checks — passed

Closes #306