optimize for local hardware

[deepfates]

Different people have different computers, and it would be nice to use the maximum available performance when indexing. Need to configure both PyTorch and Annoy based on hardware available at runtime