Add support for the timbrooks/instruct-pix2pix model

[stronk-dev]

No description provided.

[stronk-dev]

Since pix2pix has yet another parameter (image_guidance_scale), we probably want some more changes. We also need a better way to get a default inference_steps value. I think for most dApps they want a sensible default so they don't hjave to bother thinking about it, but it has so much influence in inference time and output quality. And the 'best' defaults differs depending on model and scheduler used

[stronk-dev]

Also still needs adjustment to set "fp16" automatically

[rickstaa]

Replaced by #94.

[rickstaa]

@stronk-dev I currently decided for transparency to use the models default. I however agree that in the future it would make sense to set more sensible defaults and enforce fp16. I created a linear item for it 👍🏻.