Neural Network Compression Framework (NNCF) -> VRAM saving

### Feature Idea

It would be nice to have like in sdnext as it says here [https://vladmandic.github.io/sdnext-docs/NNCF-Compression/](url), it saves lots of VRAM.

From sdnext wiki.

Features

Uses INT8, halves the model size
Saves 3.4 GB of VRAM with SDXL

Disadvantages

It is Autocast, GPU will still use 16 Bit to run the model and will be slower
Not implemented in Original backend
Fused projections are not compatible with NNCF
Using Loras will make generations slower
Options

These results compares NNCF 8 bit to 16 bit.

Model:
Compresses UNet or Transformers part of the model.
This is where the most memory savings happens for Stable Diffusion.

SDXL: 2500 MB~ memory savings.
SD 1.5: 750 MB~ memory savings.
PixArt-XL-2: 600 MB~ memory savings.

Text Encoder:
Compresses Text Encoder parts of the model.
This is where the most memory savings happens for PixArt.

PixArt-XL-2: 4750 MB~ memory savings.
SDXL: 750 MB~ memory savings.
SD 1.5: 120 MB~ memory savings.

VAE:
Compresses VAE part of the model.
Memory savings from compressing VAE is pretty small.

SD 1.5 / SDXL / PixArt-XL-2: 75 MB~ memory savings.

### Existing Solutions

_No response_

### Other

_No response_