Add option to use residual blocks for up/downsampling.

[Warvito]

Add the option to perform the downsampling and upsampling inside the Diffusion Model using Resnets Blocks (as used in here)

[Warvito]

Necessary to convert the pretrained model from "Brain Imaging Generation with Latent Diffusion Models" (https://arxiv.org/abs/2209.07162) to this package Unet model (related to #30)

[Warvito]

This component was proposed on "Diffusion Models Beat GANs on Image Synthesis" (https://arxiv.org/pdf/2105.05233.pdf), so this might be useful for other models based on this paper.