add enable sequential cpu offloading to other stable diffusion pipelines

[piEsposito]

Follow up of #1024

Adds enable_sequential_cpu_offload method to StableDiffusionInpaintPipeline and StableDiffusionImg2ImgPipeline alongs with proper tests.

Also renames the test on StableDiffusionPipeline.enable_sequential_cpu_offload to test_stable_diffusion_pipeline_with_sequential_cpu_offloading for clarity.

@patrickvonplaten here I'm just adding the enable_sequential_cpu_offload method to the img2img and inpainting pipelines for Stable Diffusion.

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.

[piEsposito]

@patrickvonplaten the tests that are not passing seem to be related to the runwayml/stable-diffusion-v1-5, which you can see I did not edit at all.

Same happened with #1010 btw.

Edit: nevermind you fixed it lightning fast.

[piEsposito]

@patrickvonplaten I've added @pedrogengo to d6fcb91 because he found the bug that this commit solves and helped me find the solution.

[pedrogengo]

@piEsposito After line 402 of inpaint pipeline file you should add:

masked_image_latents = masked_image_latents.to(device=self.device, dtype=text_embeddings.dtype)

To make sure that all tensors are on the same device.
Before add this line the inference was throwing an error for me (RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cpu and cuda:0! (when checking argument for argument tensors in method wrapper_cat)). However, after add this, everything worked fine!

Unfortunately, I could not comment direct on the file.

[piEsposito]

@patrickvonplaten thank you for the kindness of addreessing your review.

[patrickvonplaten]

Thanks a lot the PR :-)