[Modular] qwen refactor

[yiyixuxu]

This PR adds support for QwenImageLayred, along with some refactor

QwenImage-Layered

testing script

import torch
from diffusers import ModularPipeline
from diffusers.utils import load_image

device = "cuda:0"
repo_id = "Qwen/Qwen-Image-Layered"

pipeline = ModularPipeline.from_pretrained(repo_id)
pipeline.load_components(torch_dtype=torch.bfloat16)
pipeline.to(device)
image = load_image("https://huggingface.co/datasets/YiYiXu/testing-images/resolve/main/halloween.png").convert("RGBA")

output_images = pipeline(
    image=image,
    negative_prompt=" ",
    num_inference_steps=50,
    resolution=640,
    generator=torch.Generator(device="cuda").manual_seed(777),
    use_en_prompt=True,
).images[0]
for i, image in enumerate(output_images):
    image.save(f"test1_output_{i}.png")

refactors

1. Introduced ConditionalPipelineBlocks base class

Added a new base class ConditionalPipelineBlocks that allows defining custom block selection logic via select_block() method. This generalizes how conditional/branching pipelines work:

class ConditionalPipelineBlocks(ModularPipelineBlocks):
    block_classes = []
    block_names = []
    
    def select_block(self, *input_names) -> Optional[str]:
        """Custom logic to select which sub-block to run."""
        raise NotImplementedError()

AutoPipelineBlocks now becomes a specialized subclass that auto-generates select_block() from block_trigger_inputs - the previous behavior is preserved .

get_execution_blocks(*trigger_inputs) still works, so we can still pre-determine which blocks will run without actually executing the pipeline.

2. Reorganized modular blocks into pipeline-specific files

Split the monolithic modular_blocks.py into separate files per pipeline variant for better maintainability:

modular_pipelines/qwenimage/
├── modular_blocks_qwenimage.py          
├── modular_blocks_qwenimage_edit.py    
├── modular_blocks_qwenimage_edit_plus.py 
├── modular_blocks_qwenimage_layered.py  

[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]

@DN6 @sayakpaul
I merged because I want to iterate quickly on mellon, feel free to take a look when you have time later