Attempt to implement NVAE

[0xtristan]

NVAE seems to be SOTA for VAE-based models. Unlike VQ-VAE it retains the original VAE objective and is continuous.

Pytorch implementation here: https://github.com/NVlabs/NVAE

Main improvements are: Multi-scale architecture, separable convolution, swish activation function, flow model.

I think we can keep it simple and just implement a few basic improvements, ignoring more complex changes such as the addition of the flow model.