Massive overhaul of the project. It now gives just two functions and uses `hnswlib` for all the indexing. It seems like a reliable option for most use-cases.
0.1.1
Thanks to gclen you can now re-use scikit-learn pipelines without refitting them internally.