Faiss xb update and other small changes

[Gordon-BP]

The latest version of faiss depreciated the .xb attribute, causing errors when training & converting JPQ models. This PR replaces instances of:

coarse_embeds = faiss.vector_to_array(coarse_quantizer.xb)
centroid_embeds += coarse_embeds.reshape(ivf_index.pq.M, -1, ivf_index.pq.dsub)

with

coarse_embeds = faiss.rev_swig_ptr(coarse_quantizer.get_xb(), coarse_quantizer.ntotal * coarse_quantizer.d)
coarse_embeds = coarse_embeds.reshape(coarse_quantizer.ntotal, coarse_quantizer.d)

Additionally, I added a few small changes for quality of life and for running on systems without GPU:

• Added missing newline '' chars in the readme bash scripts
• Added boto3 to the requirements in setup.py
• Added behavior for a default tokenizer
• Removed the depreciating transformers AdamW optimizer and replaced it with pytorch AdamW