Deprecate upcast_vae in SDXL based pipelines

[DN6]

What does this PR do?

Removed the type cast in the AutoencoderKL Decoder in this PR in order to address graph breaks. The PR breaks some Mellon Nodes and default SDXL inference. The problem is that the casting is used to address an issue in the base SDXL VAE, which overflows when running in FP16.

The overflow issue has been fixed for a while and almost all finetunes use the fixed VAE, so this logic that conditionally upcasts the upsample layers of the VAE is no longer needed. The memory savings from this selective layer casting in also quite minimal, so we can safely upcast all layers in the VAE if needed.

This PR

1. Deprecates the upcast_vae method in the SDXL based pipelines that selectively upcasts only the upsample layers of the decoder in favour of upcasting the entire VAE if upcasting is required.

Fixes # (issue)

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline?
• Did you read our philosophy doc (important for complex PRs)?
• Was this discussed/approved via a GitHub issue or the forum? Please add a link to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

[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.