fix Mellon custom block required input handling

[wunianze666-netizen]

Summary

This PR preserves required custom block inputs when generating a Mellon config.

Motivation

While reviewing the Mellon custom block flow, I noticed that MellonPipelineConfig.from_custom_block() always emitted an empty
equired_inputs list, even when a block input was declared with
equired=True. The same code path also printed a debug line for every input.

This makes the generated Mellon schema less accurate for custom blocks and can leak noisy stdout from a library API. This came up while reading through the modular Mellon findings in #13650.

Changes

• track required custom block inputs in rom_custom_block()
• stop printing a debug line for each processed input
• add a regression test covering the generated
  equired_inputs metadata

Testing

• python -m py_compile src/diffusers/modular_pipelines/mellon_node_utils.py tests/modular_pipelines/test_modular_pipelines_custom_blocks.py

Notes

I could not run the pytest target in this local environment because pytest and the repo's development dependencies are not installed here, so I limited verification to a syntax check and a focused regression test addition.

[yiyixuxu]

thanks

[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.

[wunianze666-netizen]

Fixed the modular CI regression in 6ab3e4d: required_inputs is retained in the internal node spec and consumed when building node_params, but it is not part of the public node_params output. The regression test now checks the spec-level metadata and the resulting required label separately. After updating the branch to current main, the focused test passes locally on Windows (1 passed in 3.91s).