Align DyPE with the paper as much as possible

[JPPhoto]

Summary

Reworks FLUX DyPE to behave more like the paper while preserving the existing invocation surface and preset wiring.

This PR does three things:

• Aligns the core DyPE math with a paper-style scheduler by replacing the previous repo-specific RoPE scaling heuristic with a direct timestep scheduler in the active vision_yarn path.
• Keeps DyPE correctly synchronized with the actual denoiser noise schedule across all Flux scheduler paths, including the built-in Euler path and diffusers euler, heun, and lcm.
• Retunes the area preset so it is no longer overly aggressive with the simpler paper-like DyPE core, which reduces artifact risk on larger and more extreme aspect-ratio outputs.

The implementation stays compatible with the current invocation/UI fields. Internally, the unused DyPE branches were removed, the remaining runtime path was simplified, and tests were added/updated to cover scheduler integration and preset behavior.

Related Issues / Discussions

Attempts to resolve ongoing DyPE issues related to loss of high-resolution detail, stretching, etc.

QA Instructions

Run the targeted tests:

• pytest tests/backend/flux/dype/test_dype.py
• pytest tests/backend/flux/test_denoise.py
• pytest tests/app/invocations/test_flux_denoise.py

Test manually:

1. Generate FLUX images with DyPE enabled using:
   • scheduler=euler
   • scheduler=heun
   • scheduler=lcm
   • and the built-in Euler path
2. Compare dype_preset=auto vs. dype_preset=area at different resolutions:
   • 1536x1024
   • 2048x1152
   • 2304x960
3. Confirm:
   • DyPE runs without scheduler-related errors
   • area is closer to auto in strength than before
   • ultra-wide results no longer use the previous overly aggressive area-based settings

Merge Plan

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
• Documentation added / updated (if applicable)
• Updated What's New copy (if doing a release after this PR)

[JPPhoto]

This still suffers from high-frequency detail loss that probably is going to be a side effect of DyPE. I don't think we'll be able to fix that without veering far from the original implementation again.