Elasticsearch HNSW hits 0.99 recall at 15,000 QPS

Elasticsearch HNSW achieves 0.99 recall@10 at 15,000 QPS on float32 vectors. With DiskBBQ quantization, recall drops to 0.97 at 55,000 QPS, a 3.7x throughput gain for a 0.02 recall reduction. KD-trees fail above 20 dimensions due to the curse of dimensionality, making HNSW necessary for high-dimensional ANN search.