[Modular] add explicit workflow support

[yiyixuxu]

This PR adds support for explicitly defining and retrieving workflows (currently only added to SequentialPipelineBlocks).

Previously, to make sure users understand what workflows a pipeline supports, we:

• we try to made notes about it in description property whenever we can (example https://github.com/huggingface/diffusers/blob/main/src/diffusers/modular_pipelines/qwenimage/modular_blocks_qwenimage.py#L475)
• we provided get_execution_blocks method for user to understand code path will execute based on specific inputs https://github.com/huggingface/diffusers/blob/main/src/diffusers/modular_pipelines/modular_pipeline.py#L1059
• we also try to maintained a list of all the blocks for all workflows in ALL_BLOCKS (example: flux/modular_blocks.py#L440) with TEXT2IMAGE_BLOCKS, IMAGE_CONDITIONED_BLOCKS, etc.

WIth this PR, workflows are now:

• explicitly defined via _workflow_map on the auto blocks class, see qwen example

class QwenImageAutoBlocks(SequentialPipelineBlocks):
    ...
    _workflow_map = {
        "text2image": {"prompt": True},
        "image2image": {"prompt": True, "image": True},
        "inpainting": {"prompt": True, "mask_image": True, "image": True},
        "controlnet_text2image": {"prompt": True, "control_image": True},
        "controlnet_image2image": {"prompt": True, "image": True, "control_image": True},
        "controlnet_inpainting": {"prompt": True, "mask_image": True, "image": True, "control_image": True},
    }
   ...

• Auto-documented in the class docstring (in qwen example with the _workflow_map we defined above, the auto docstring look like this)

# auto_docstring
class QwenImageAutoBlocks(SequentialPipelineBlocks):
    """
    Auto Modular pipeline for text-to-image, image-to-image, inpainting, and controlnet tasks using QwenImage.

      Supported workflows:
        - `text2image`: requires `prompt`
        - `image2image`: requires `prompt`, `image`
        - `inpainting`: requires `prompt`, `mask_image`, `image`
        - `controlnet_text2image`: requires `prompt`, `control_image`
        - `controlnet_image2image`: requires `prompt`, `image`, `control_image`
        - `controlnet_inpainting`: requires `prompt`, `mask_image`, `image`, `control_image`

      Components:
          ...

      Inputs:
          ...

      Outputs:
          ...
    """

• Programmatically accessible via workflow_names and get_workflow()
  for example, to get the "controlnet_text2image" workflow from qwen

from diffusers import ModularPipeline
pipe = ModularPipeline.from_pretrained("Qwen/Qwen-Image")
text2image_blocks = pipe.blocks.get_workflow("controlnet_image2image"")
print(text2image_blocks.sub_blocks)

InsertableDict([
  0: ('text_encoder', <obj 'diffusers.modular_pipelines.qwenimage.encoders.QwenImageTextEncoderStep'>),
  1: ('controlnet_vae_encoder', <obj 'diffusers.modular_pipelines.qwenimage.encoders.QwenImageControlNetVaeEncoderStep'>),
  2: ('denoise.input', <obj 'diffusers.modular_pipelines.qwenimage.inputs.QwenImageTextInputsStep'>),
  3: ('denoise.controlnet_input', <obj 'diffusers.modular_pipelines.qwenimage.inputs.QwenImageControlNetInputsStep'>),
  4: ('denoise.prepare_latents', <obj 'diffusers.modular_pipelines.qwenimage.before_denoise.QwenImagePrepareLatentsStep'>),
  5: ('denoise.set_timesteps', <obj 'diffusers.modular_pipelines.qwenimage.before_denoise.QwenImageSetTimestepsStep'>),
  6: ('denoise.prepare_rope_inputs', <obj 'diffusers.modular_pipelines.qwenimage.before_denoise.QwenImageRoPEInputsStep'>),
  7: ('denoise.controlnet_before_denoise', <obj 'diffusers.modular_pipelines.qwenimage.before_denoise.QwenImageControlNetBeforeDenoiserStep'>),
  8: ('denoise.controlnet_denoise', <obj 'diffusers.modular_pipelines.qwenimage.denoise.QwenImageControlNetDenoiseStep'>),
  9: ('denoise.after_denoise', <obj 'diffusers.modular_pipelines.qwenimage.decoders.QwenImageAfterDenoiseStep'>),
  10: ('decode.decode', <obj 'diffusers.modular_pipelines.qwenimage.decoders.QwenImageDecoderStep'>),
  11: ('decode.postprocess', <obj 'diffusers.modular_pipelines.qwenimage.decoders.QwenImageProcessImagesOutputStep'>)
])

I will add a test suit for this too!

[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.

[yiyixuxu]

TODO: update InputParams/OutputParams to remove the TODO from doc strings

[yiyixuxu]

TODO: add a test the output from workflow blocks against the auto blocks

[yiyixuxu]

TODO: support set