ext-vector: HNSW vector index with SIMD distance kernels.
Provides approximate nearest neighbor search through three procedures:
vector.build(dim, metric?)→ initializes HNSW index (metric: "l2" or "cosine")vector.add(id, vector_csv)→ inserts a vector, returns{status: STRING}vector.search(query_csv, k)→ ANN results{id: INT64, distance: DOUBLE}
Uses NEON intrinsics on aarch64 and scalar fallback (auto-vectorized to AVX2) elsewhere. Staging buffer accumulates vectors before bulk-inserting into the live HNSW index.