Binary and Scalar Embedding Quantization for Significantly Faster & Cheaper Retrieval