Function knn_search 

pub fn knn_search(
    conn: &Connection,
    embedding: &[f32],
    namespace: &str,
    k: usize,
) -> Result<Vec<(i64, f32)>, AppError>

Searches the entity_embeddings table for the k nearest neighbours using pure-Rust cosine similarity.

v1.0.76: sqlite-vec was removed. The full table scan + in-process cosine is O(N × D) per call. For namespaces with more than ~10k entities, the operator should rely on FTS5 (hybrid-search) for coarse filtering before reaching this function.

Errors

• AppError::Database — SQLite query failure.
• AppError::Embedding — invalid or mismatched embedding dimension.