fix(z_image): use unrestricted image self-attention for regional prompting

[Pfannkuchensack]

Summary

Changes image self-attention from restricted (region-isolated) to unrestricted (all image tokens can attend to each other), similar to the FLUX approach.

This fixes the issue where ZImage-Turbo with multiple regional guidance layers would generate two separate/disconnected images instead of compositing them into a single unified image.

The regional text-image attention remains restricted so that each region still responds to its corresponding prompt.

Fixes #8715

Related Issues / Discussions

• Issue [bug]: ZImage-Turbo regional guidance not working as expected #8715: ZImage-Turbo regional guidance not working as expected

QA Instructions

1. Create a new canvas generation with ZImage-Turbo
2. Add two or more regional guidance layers with different prompts (e.g., "woman in business suit" on the left, "woman in peasant dress" on the right)
3. Add a global prompt (e.g., "two women")
4. Generate the image
5. Expected: A single unified image with both women composited according to their regions
6. Previous behavior: Two separate, independent images

Merge Plan

No special merge considerations required. This is a targeted fix to the attention mask construction.

Checklist

• The PR has a short but descriptive title, suitable for a changelog
• Tests added / updated (if applicable)
• ❗Changes to a redux slice have a corresponding migration - N/A
• Documentation added / updated (if applicable) - N/A
• Updated What's New copy (if doing a release after this PR)

[lstein]

This change works well in my hands. I find that even minimal overlap of the regions ensures better coherence to the main prompt, but I usually get acceptable results even when the regions are not overlapping.