[SD-XL] Flax implementation

[mar-muel]

What does this PR do?

Adds Flax implementation for SD-XL pipeline 🚀

Related issue: #4007

This is still WIP unfortunately 😅

• I've managed to convert the weights using from_pt=True, however this might currently not work out of the box for
  text_encoder_2 (as transformers currently doesn't support CLIPTextModelWithProjection for Flax). I've pushed my version of the converted flax weights to this repo for now: https://huggingface.co/nyxai-lab/sdxl-0.9-flax/tree/main
• I've confirmed forward passes for UNet and VAE to work correctly. It seems the text encoders currently yield correct results for hidden states, but still getting different results for pooled layers EDIT: Upon checking again the outputs seem close enough
• Refiner stage and part of txt2img is still WIP
• I haven't looked into safety checker, watermark, docs or tests yet

Hope this can serve as a starting point :)

EDIT: Adding some outputs here (based on this PR):

a beautiful stack of rocks sitting on top of a beach, a picture, red black white golden colors, chakras, packshot, stock photo

photo of a rhino dressed suit and tie sitting at a table in a bar with a bar stools, award winning photography, Elke vogelsang

[pcuenca]

Awesome @mar-muel, many thanks! We'll take a look soon!

[pcuenca]

Suggested change

	for tokenizer in enumerate(tokenizers):
	text_inputs = self.tokenizer(
	for tokenizer in tokenizers:
	text_inputs = tokenizer(

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint.

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.