Support for Flux 1.58 bit model

### Feature Idea

There's a new paper from ByteDance where they quantise the flux model with 1.58 bits per parameter. The code is yet to be released, but it would be good to have for low v-ram users. But it would need a custom kernel. Not sure if it makes sense to add to the core.

Project page: https://chenglin-yang.github.io/1.58bit.flux.github.io/

### Existing Solutions

_No response_

### Other

_No response_