Hybrid GPU–CPU Approach to Faster Vector Indexing and Cheaper Queries