Build and query small approximate nearest neighbour (ANN) indices on the JVM.
This focuses on building small ANN indices of large word (or other entity) embeddings that can be loaded into memory. It is designed to work with O(10m) vectors with tens, hundreds or thousands of dimensions and return queries in 5ms or less on modern hardware with sane defaults.