feat: expand unifiedSearch to accept object input with optional vector (#855)

[pyramation]

Summary

Resolves the unifiedSearch input modality mismatch (#855). Previously, unifiedSearch accepted a plain String that could only be dispatched to text-compatible adapters (tsvector, BM25, trgm). pgvector couldn't participate because it needs a vector array, not text.

Before:

allDocuments(where: {
  unifiedSearch: \"machine learning\"          # text only
  vectorEmbedding: { vector: [0.12, ...] }   # separate filter, AND'd
})

After:

# Text-only (new object syntax)
allDocuments(where: { unifiedSearch: { text: \"machine learning\" } })

# Vector-only
allDocuments(where: { unifiedSearch: { vector: [0.12, ...], metric: COSINE } })

# True hybrid (text OR vector, OR-combined)
allDocuments(where: {
  unifiedSearch: {
    text: \"machine learning\"
    vector: [0.12, ...]
    metric: COSINE
    distance: 1.5
  }
})

Changes:

• New UnifiedSearchInput GraphQL input type registered in the init hook with fields: text, vector, metric, distance, includeChunks
• unifiedSearch filter field type changed from String to UnifiedSearchInput
• Text path dispatches to all text-compatible adapters (tsvector, BM25, trgm) — same as before
• Vector path dispatches to pgvector adapter — new
• WHERE clauses from all active paths are OR-combined (true hybrid retrieval)
• searchScore blends all active signals (text + vector) into a single 0..1 relevance number
• Vector fields (vector, metric, distance, includeChunks) only appear when a pgvector adapter is registered
• Hook point for future Graphile LLM plugin: when vector is omitted and an LLM plugin is loaded, it can intercept the text input and auto-generate the vector before the search plugin processes it

Breaking change: unifiedSearch no longer accepts a plain string. Callers must use unifiedSearch: { text: \"...\" } instead of unifiedSearch: \"...\".

Review & Testing Checklist for Human

Medium risk — API shape change for existing unifiedSearch consumers.

• Verify any existing code using unifiedSearch: \"text\" is updated to unifiedSearch: { text: \"text\" }
• Test hybrid query with both text and vector — verify rows matching EITHER path are returned (OR semantics)
• Test vector-only query via unifiedSearch: { vector: [...], distance: 1.0 } — verify vector distance scoring works
• Test that searchScore correctly blends text + vector signals when both are active
• Verify includeChunks option works for vector search through unifiedSearch input

Notes

• The UnifiedSearchInput type conditionally includes vector fields — only when a pgvector adapter is registered. Without pgvector, only text is available.
• The existing per-adapter filter fields (tsvTsv, bm25Body, vectorEmbedding, etc.) continue to work independently and can be combined with unifiedSearch.
• This is a plugin-only change — no database-side changes needed.

Link to Devin session: https://app.devin.ai/sessions/2b5a29d83d3f478e8d3d972653b4879c
Requested by: @pyramation

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment and CI monitoring

[blacksmith-sh]

Found 59 test failures on Blacksmith runners:

Failures

Test	View Logs
BM25 search (pg_textsearch)/BM25 filter field exists on LocationFilter	View Logs
BM25 search (pg_textsearch)/BM25 orderBy sorts by relevance	View Logs
BM25 search (pg_textsearch)/BM25 score field is populated when filter is active	View Logs
BM25 search (pg_textsearch)/BM25 score is null when no BM25 filter is active	View Logs
BM25 search (pg_textsearch)/BM25 search filters by body text	View Logs
CLI E2E — search commands against real DB/ should combine search filter with --limit for paginated results	View Logs
CLI E2E — search commands against real DB/ should expose search fields on Article type via introspection	View Logs
CLI E2E — search commands against real DB/ should filter articles by trgm similarity via dot-notation where	View Logs
CLI E2E — search commands against real DB/ should filter articles by tsvector search via --where.tsvTsv	View Logs
CLI E2E — search commands against real DB/ should filter via unifiedSearch composite filter	View Logs
Kitchen sink (multi-plugin queries)/combines BM25 + scalar filter	View Logs
Kitchen sink (multi-plugin queries)/combines OR with tsvector and scalar filters	View Logs
Kitchen sink (multi-plugin queries)/ combines tsvector search + scalar filter + relation filter	View Logs
Kitchen sink (multi-plugin queries)/ mega query: BM25 + tsvector + pgvector + PostGIS + pg_trgm + relation filter + scalar i n ONE query, with multi-signal orderBy	View Logs
Kitchen sink (multi-plugin queries)/pagination works with filters	View Logs
Kitchen sink (multi-plugin queries)/vector search + filter on results	View Logs
Many-to-many type name collision resilience/ opted-in junction tables produce m2m types in the schema	View Logs
Many-to-many type name collision resilience/ produces distinct m2m edge types for each junction path	View Logs
Many-to-many type name collision resilience/ schema builds without crashing when two junction tables target the same pair	View Logs
pgvector/exposes embedding as array of floats	View Logs
pgvector/vector search function returns results ordered by similarity	View Logs
pgvector/vector search respects result_limit	View Logs
PostGIS spatial filters/geom column is exposed as GeoJSON	View Logs
Relation filters/backward relation exists: locations that have tags	View Logs
Relation filters/backward relation none: categories with NO inactive locations	View Logs
Relation filters/backward relation some: categories with at least one active location	View Logs
Relation filters/combines relation filter with scalar filter	View Logs
Relation filters/forward relation: filter locations by category name	View Logs
Scalar and logical filters/filters by boolean field	View Logs
Scalar and logical filters/filters by numeric comparison	View Logs
Scalar and logical filters/filters by string equalTo	View Logs
Scalar and logical filters/isNull filters for NULL rating	View Logs
Scalar and logical filters/no filter returns all rows	View Logs
Scalar and logical filters/NOT negates a condition	View Logs
Scalar and logical filters/OR combines conditions	View Logs
Schema introspection/Location type has vector embedding field	View Logs
Schema introspection/LocationFilter has relation filter fields	View Logs
Schema introspection/LocationFilter has tsvector search field	View Logs
Schema introspection/LocationFilter type exists and has scalar filter fields	View Logs
Schema introspection/ locations connection exists and has where argument but no condition	View Logs
tsvector search (PgSearchPlugin)/tsvector search combined with scalar filter	View Logs
tsvector search (PgSearchPlugin)/tsvTsv matches filters by text search	View Logs
tsvector search (PgSearchPlugin)/tsvTsv with broad term matches multiple rows	View Logs
Unified Search — server integration/should query all articles	View Logs
Unified Search — server integration › composite search/ should expose searchScore when any search filter is active	View Logs
Unified Search — server integration › composite search/ should filter via unifiedSearch composite filter	View Logs
Unified Search — server integration › composite search/ should order by SEARCH_SCORE_DESC	View Logs
Unified Search — server integration › composite search/ should return searchScore as null when no search filter is active	View Logs
Unified Search — server integration › Mega Query v1 — per-algorithm filters/ should combine tsvector + trgm filters with multi-column ordering	View Logs
Unified Search — server integration › Mega Query v2 — unifiedSearch composite/ should use unifiedSearch + SEARCH_SCORE_DESC	View Logs
Unified Search — server integration › pg_trgm fuzzy matching/ should filter articles by trgm similarity on title (trgmTitle)	View Logs
Unified Search — server integration › pg_trgm fuzzy matching/ should filter by trgm on body (trgmBody)	View Logs
Unified Search — server integration › pg_trgm fuzzy matching/ should order by TITLE_TRGM_SIMILARITY_DESC	View Logs
Unified Search — server integration › Schema introspection/ should expose expected filter fields on ArticleFilter type	View Logs
Unified Search — server integration › Schema introspection/ should expose expected orderBy enum values	View Logs
Unified Search — server integration › Schema introspection/ should expose expected search fields on Article type	View Logs
Unified Search — server integration › tsvector search/ should filter articles by tsvector search (tsvTsv)	View Logs
Unified Search — server integration › tsvector search/should order by TSV_RANK_DESC	View Logs
Unified Search — server integration › tsvector search/ should return tsvRank as null when no tsvector filter is active	View Logs