Fix: Ensure Consistent MASK Output Dimensions from ReduxAdvanced Node

[sko00o]

Description:

This PR addresses an inconsistency in the dimensions of the MASK tensor returned by the ReduxAdvanced node.

Problem:

Previously, the shape of the returned masko tensor depended on the mode setting, specifically on whether the internal prepareImageAndMask function utilized the crop or letterbox helper function:

1. crop path (mode="center crop (square)"): This path (after recent user edits) produced a mask with shape (batch, height, width, 1).
2. letterbox path (mode="keep aspect ratio" or "autocrop with mask"): This path produced a mask with shape (batch, 1, height, width).

The final return statement in apply_stylemodel used masko.squeeze(-1). This correctly processed the (b, h, w, 1) output from crop into the standard (b, h, w) mask format. However, it did not correctly process the (b, 1, h, w) output from letterbox, leaving it as a 4D tensor (unless width happened to be 1).

This inconsistency meant that the MASK output type did not have a stable dimension format, potentially causing issues with downstream nodes that expect the standard ComfyUI 3D mask format (batch, height, width), such as Conditioning (Set Mask) or GrowMask. While the MaskToImage node could handle both formats, this masked the underlying inconsistency in the ReduxAdvanced node's output.

Solution:

The core change is minimal and focuses on unifying the output format before the final squeeze.

1. Modified letterbox function: The final step in handling the mask within letterbox was changed from .view(b, 1, desiredSize, desiredSize) (which was effectively a no-op) to .movedim(1, -1). This moves the channel dimension (dim 1) to the end, transforming the shape from (b, 1, h, w) to (b, h, w, 1).
2. Consistent Output: With this change, both crop and letterbox now return masks with the shape (b, h, w, 1) (when a mask is present).
3. Effective Squeeze: Consequently, the existing masko.squeeze(-1) in the apply_stylemodel's return statement now correctly handles the output from both paths, reliably producing the standard 3D mask format (batch, height, width).

Benefit:

This change guarantees that the ReduxAdvanced node's MASK output is always a standard 3D tensor (batch, height, width), improving compatibility and predictability when connecting to other ComfyUI nodes.