Differentiable VQ-VAE's for Robust White Matter Streamline Encodings