when using kmeans init, initialize cluster sizes too

use threshold of ema cluster sizes to determine which codes to replac…

…e, as in soundstream paper, thanks to @wesbz

add code expiry / replacement strategy from soundstream paper

add ability to use cosine similarity for measuring distance to codes

add measure number one for dead codebooks

add kmeans init of codebook, as proposed in soundstream paper, also m…

…ake sure commitment loss is not calculated on eval

allow for access to codebook with property method

make init param more intuitive, but leave n_embed backwards compatible

add residual vector quantization, from soundstream paper from google