feat: implement lexical search functionality in RAG, including new SQL queries and service methods

db/migrations/0000018_rag_chunks_content_trgm.down.sql

@@ -0,0 +1 @@
+DROP INDEX IF EXISTS idx_rag_chunks_content_trgm;

db/migrations/0000018_rag_chunks_content_trgm.up.sql

@@ -0,0 +1,4 @@
+CREATE EXTENSION IF NOT EXISTS pg_trgm;
+
+CREATE INDEX IF NOT EXISTS idx_rag_chunks_content_trgm
+ON rag_chunks USING GIN (content gin_trgm_ops);

db/query/rag.sql

@@ -19,6 +19,42 @@ WHERE 1=1
 ORDER BY embedding <=> sqlc.arg('embedding')
 LIMIT sqlc.arg('match_count');
 
+-- name: SearchRagChunksLexical :many
+WITH terms AS (
+    SELECT DISTINCT trim(term) AS term
+    FROM unnest(sqlc.arg('terms')::text[]) AS term
+    WHERE trim(term) <> ''
+),
+matched AS (
+    SELECT
+        r.id,
+        r.source_type,
+        r.source_id,
+        r.project_id,
+        r.chunk_index,
+        r.content,
+        r.created_at,
+        r.updated_at,
+        COUNT(*)::float8 AS similarity
+    FROM rag_chunks r
+    JOIN terms t ON r.content ILIKE '%' || t.term || '%'
+    WHERE (sqlc.narg('project_id')::uuid IS NULL OR r.project_id = sqlc.narg('project_id')::uuid)
+    GROUP BY
+        r.id,
+        r.source_type,
+        r.source_id,
+        r.project_id,
+        r.chunk_index,
+        r.content,
+        r.created_at,
+        r.updated_at
+)
+SELECT
+    id, source_type, source_id, project_id, chunk_index, content, created_at, updated_at, similarity
+FROM matched
+ORDER BY similarity DESC, chunk_index ASC
+LIMIT sqlc.arg('match_count');
+
 -- name: DeleteRagChunksBySourceIDs :exec
 DELETE FROM rag_chunks
 WHERE source_type = $1 AND source_id = ANY($2::uuid[]);