Init porting VQ-diffusion to diffusers

This initial commit ports the VQ-diffusion VQVAE for the ITHQ dataset to
diffusers.

Add `convert_vq_diffusion_to_diffusers.py` script:
This script initially only converts the VQVAE to diffusers. It will be
updated to convert the whole model.

Add placeholder `VQDiffusionPipeline`:
The `VQDiffusionPipeline` is added as a placeholder to wrap the vqvae so
it can be used in the `convert_vq_diffusion_to_diffusers.py` script to
save the ported model.

Add `ConvAttentionBlock`:
The VQVAE used for ITHQ in VQ-diffusion uses a slightly different
attention block than the one already in diffusers. The
`ConvAttentionBlock` uses `torch.nn.Conv2d`'s for its linear layers
as opposed to `torch.nn.Linear`'s. There are a few other minor
discrepancies between the two attention blocks.

Add specify dimmension of embeddings to VQModel:
`VQModel` will by default set the dimension of embeddings to the number
of latent channels. The VQ-diffusion VQVAE for ITHQ has a smaller
embedding dimension, 128, than number of latent channels, 256.