Add tutorial to perform super-resolution with Latent Diffusion Models

[Warvito]

Create tutorial that performs super-resolution based on the upscaler networks of the stable diffusion v2.0 (https://stability.ai/blog/stable-diffusion-v2-release).
This would include to use the the noise augmentation technique that includes adding noise to the lower resolution image as described in the paper from Imagen Video (Section 2.5) https://imagen.research.google/video/paper.pdf . The variance of the added noise is defined by the noise_level which is also passed to the network as class conditioning. During inference the same process is used as show in the huggingface diffuser code (https://github.com/huggingface/diffusers/blob/21e61eb3a9d16a46245bd284fea3aa19e66772f5/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_upscale.py#L426) and stability.ai code ( https://github.com/Stability-AI/stablediffusion/blob/d55bcd4d31d0316fcbdf552f2fd2628fdc812500/scripts/gradio/superresolution.py#L52)

[Warvito]

Original-code to train the upscaler models available at
https://github.com/Stability-AI/stablediffusion/blob/d55bcd4d31d0316fcbdf552f2fd2628fdc812500/ldm/models/diffusion/ddpm.py#L1352