[core] feat: implement automodel for diffusers.

[sayakpaul]

What does this PR do?

Fixes #10059.

Amidst such a growing number of model classes, I think it makes sense for diffusers to have its AutoModel class.

Instead of defining a list of models manually, I have taken the approach of using importlib to directly import the model from a loaded config. I think this is the simplest and easiest but I am not too fixated on this.

I am opening this PR to gather feedback and iterate. Letting Dhruv to do that first and then will request reviews from others.

I haven't added tests (can do) for it but AutoModel also works with quantization i.e.,:

from diffusers.models.auto_model import AutoModel

model = AutoModel.from_pretrained("hf-internal-testing/flux.1-dev-nf4-pkg", subfolder="transformer")
print(model.__class__.__name__)

Successfully resolves to FluxTransformer2DModel.

TODO

• Add AutoModel to init.
• Docs
• Additional tests if needed (I think I have covered the crucial ones)

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.