[Cosmos3] Fix Transfer SeaCache artifacts with control CFG

[yzhautouskay]

What does this PR do?

Follow-up to #14663

Fixes Cosmos3 Transfer quality artifacts when SeaCache is combined with control CFG. The cache indicator now uses only noisy target latents, excluding separate clean control-hint latents. Control hints still condition the transformer, and each CFG branch retains its own residual history.

Validation:

• Added regressions for target-only indicators, branch histories, cache decisions, residual orders 0/1, and state reset
• All six BF16/FP8 comparisons improved LPIPS, PSNR, and SSIM while retaining 2.10-2.18× speedup
• I2V, T2V, and action modes were bitwise identical before and after the change in BF16 and FP8
• 47 model tests and 19 targeted SeaCache tests passed; make style and make fix-copies passed

Self-review: READY. No blocking, non-blocking, or dead-code findings.

Test results

The current cache quality issue is input-dependent because control-CFG branch schedules can diverge very differently across examples, independent of the control type alone. Some show little degradation, while others develop severe temporal/block artifacts. The effect is amplified with FP8-quantized checkpoints.

This PR addresses the observed flickering and severe degradation in both BF16 and FP8 runs. All six example/precision comparisons improve LPIPS, PSNR, and SSIM against their matching uncached baseline.

Rows: BF16 (top), FP8 (bottom).
Columns, left to right: uncached / current cache / this PR fix.
Videos include latency, speedup, LPIPS, PSNR, and SSIM.

diffusers_0200_edge_six_arm.mp4

diffusers_0539_seg_six_arm.mp4

Before submitting

• Did you use an AI agent (Claude Code, Codex, Cursor, etc.) to help with this PR? If so:
  • Did you read the Coding with AI agents guide?
  • Did you run the self-review skill on the diff?
  • Did you share the final self-review notes in the PR description or a comment?
• Did you read the contributor guideline?
• Did you read our philosophy doc? (important for complex PRs)
• Was this discussed/approved via a GitHub issue or the forum? Please add a link to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?
• Are you the author (or part of the team) of the model/pipeline (only applicable for model/pipeline related PRs)?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.