Semantic search across the Bible (with UB paragraphs)

Free-form natural-language search across all 17,641 Bible chunks. Each result includes the top-N pre-computed Urantia paragraphs related to that chunk via the existing cross-reference data, so a single query surfaces both the Bible matches and the relevant UB content.

Query is embedded via text-embedding-3-small (1536-d) and matched against bible_chunks.embedding_small with a pgvector HNSW index. Latency is ~50-100ms on cache miss for the embedding call, ~30ms cached.

Optional filters: canon (ot, deuterocanon, nt), bookCode (any OSIS/USFM/full-name/alias). urantiaParallelLimit controls how many UB paragraphs to attach per result (0-10, default 3). Set to 0 to suppress.