fix(image_processor): maintain 3-tuple return contract in InpaintProcessor.preprocess when mask is None

[patrickswedish]

What does this PR do?

Fixes #14470.

InpaintProcessor.preprocess returns a 3-tuple (image, mask, postprocessing_kwargs) across all masked execution paths, but when mask is None, the early return returned only the processed image tensor directly. Callers unpacking the standard 3-value contract (image, mask, postprocessing_kwargs = processor.preprocess(...)) would raise ValueError: not enough values to unpack (expected 3, got 1).

Root Cause

When the 3-value contract and postprocessing_kwargs were introduced in commit f50b18eec (#12220), the early return for mask is None was not updated to return the consistent 3-tuple structure. Additionally, the return type annotation indicated tuple[torch.Tensor, torch.Tensor], which did not reflect the 3-element return value or optional mask.

Changes

• Updated InpaintProcessor.preprocess in src/diffusers/image_processor.py when mask is None to return (processed_image, None, postprocessing_kwargs) where postprocessing_kwargs is populated with {"crops_coords": None, "original_image": None, "original_mask": None}.
• Updated return type annotation to tuple[torch.Tensor, torch.Tensor | None, dict[str, Any]].
• Added unit tests in tests/others/test_image_processor.py covering preprocess with mask, without mask (mask=None), and with padding_mask_crop.

Before submitting

• Did you read the contributor guideline?
• Did you read our philosophy doc?
• Was this discussed/approved via a GitHub issue or the forum? Fixes InpaintProcessor.preprocess returns a bare tensor when no mask is passed, breaking its own 3-value contract #14470.
• Did you write any new necessary tests?

Who can review?

@DN6 @yiyixuxu

[lindicaphxag-tech]

I closed my duplicate #14807 in favor of this PR after comparing the diffs. This change covers the same no-mask contract and also tests masked and padding-crop paths with the precise optional-mask return type. Could a maintainer review or identify any remaining concern?

[yiyixuxu]

can we just reduce to one test? you can use @pytest.mark.parametrize

[patrickswedish]

Thanks for the feedback, @yiyixuxu! Updated to consolidate into a single @pytest.mark.parametrize test covering the with-mask, without-mask, and crop configurations. Rebased cleanly on latest main, squashed to a single atomic commit, and verified all tests and linting pass locally.