[Euler Discrete] do sigmas edits from the pipeline.

[sayakpaul]

What does this PR do?

Fixes the issue for which #6078 was made.

[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.

[DN6]

Pipeline is breaking with LCM
https://github.com/huggingface/diffusers/actions/runs/7123821250/job/19397082791?pr=6081#step:11:46

[sayakpaul]

@DN6 could you give this another look?

[sayakpaul]

@patrickvonplaten @DN6

Since the concerned change is very specific to the torch.compile() performance how about we do from the pipeline calling script?

pipe.scheduler.sigmas = pipe.scheduler.sigmas.tolist()

This essentially solves the purpose. In retrospect, I think we shouldn't touch the internals and instead leverage what we can to keep changes more explicit.

[sayakpaul]

Closing because we can simply do: pipeline.scheduler.sigmas = pipeline.scheduler.sigmas.cpu() from the pipeline calling script.