PostgreSQL + pgvector backend for mnemo (StorageBackend + VectorIndex).
Capability note — semantic recall is supported (v0.5.7, #99). Embeddings
are persisted to the pgvector vector column and, when [PgVectorIndex] is
constructed with a pool ([PgVectorIndex::with_pool]), semantic / auto
(hybrid) / graph / domain_scoped / reconstruct recall run a real
cosine-distance ANN query against the idx_memories_embedding_hnsw HNSW
index. The synchronous VectorIndex trait
is bridged to async sqlx with block_in_place + Handle::block_on, so the
Postgres vector path requires the multi-threaded Tokio runtime (the
CLI/server is #[tokio::main]). filtered_search uses the same
permission-safe oversample-then-filter as the USearch backend.
If the pgvector extension / <=> operator is genuinely absent at runtime, or
the index is built without a pool, the vector path still fails loud with a
typed Error::BackendUnsupported
(backend = "postgres", capability = "semantic_recall") — never a silent
empty result. The long-term async-VectorIndex refactor is tracked at
https://github.com/sattyamjjain/mnemo/issues/99.