Add AttentionMixin to transformers that are missing it

[HaozheZhang6]

What does this PR do?

Several transformers build their attention with AttentionModuleMixin but don't inherit the model-level AttentionMixin, so attn_processors / set_attn_processor / fuse_qkv_projections / unfuse_qkv_projections raise AttributeError. This adds AttentionMixin to all of them, the same way WanVACETransformer3DModel was handled (#12186):

• OvisImageTransformer2DModel (the original ovis_image model/pipeline review #13630 finding)
• BriaTransformer2DModel, BriaFiboTransformer2DModel
• AnyFlowTransformer3DModel, AnyFlowFARTransformer3DModel
• ErnieImageTransformer2DModel
• LongCatAudioDiTTransformer

The attention modules already use AttentionModuleMixin, so the four APIs work without further changes and resolve to each model's own processor. Each model gets a regression test that attn_processors is populated and set_attn_processor round-trips.

Note: AnyFlowTransformer3DModel, AnyFlowFARTransformer3DModel, and ErnieImageTransformer2DModel have some unrelated pre-existing failures in their model test suites on main (not touched here); the new test_attention_processor_api passes for all seven.

Addresses the AttentionMixin pattern from #13656.

Before submitting

• Did you use an AI agent (Claude Code, Codex, Cursor, etc.) to help with this PR? If so:
  • Pointed it at .ai/ — read AGENTS.md and review-rules.md.
  • Self-reviewed the diff against .ai/review-rules.md.
• Did you read the contributor guideline?
• Was this discussed via a GitHub issue? Yes — [meta issue] Systematic model/pipeline review findings / tracking #13656 / ovis_image model/pipeline review #13630.
• Did you write any new tests?

Who can review?

@hlky — batched per your suggestion, from your model reviews in #13656.

[HaozheZhang6]

While here I checked the other transformers that build their attention with AttentionModuleMixin but don't inherit the model-level AttentionMixin — the same gap is in AnyFlowTransformer3DModel, AnyFlowFARTransformer3DModel, BriaTransformer2DModel, BriaFiboTransformer2DModel, and ErnieImageTransformer2DModel (all raise AttributeError on attn_processors). Happy to fix those in a follow-up batch if this approach looks right.

[hlky]

@HaozheZhang6 If you can identify a pattern from all the issues, then it is better to fix those all at once. AttentionMixin is a good case for that as the change is minimal. Other patterns where the change is more involved may be riskier to fix as a whole. There may be others beyond than those you found, I would either point your agent at the meta issue ask it to check all the sub-issues, or do another pass on repo as a whole. For this particular pattern the part I would exercise caution with is the tests, depending on maintainers view (I would ask @DN6 or @sayakpaul about this) it may be better to add the test to ModelTesterMixin rather than individually.

[HaozheZhang6]

Makes sense — I'll consolidate them into this PR. I'll re-scan for the full set (including any beyond the five I listed) and add AttentionMixin + a regression test to each here, then close the separate Bria PR I'd just opened (#13943).

[HaozheZhang6]

Done — all seven are in now: OvisImage, Bria, BriaFibo, AnyFlow, AnyFlowFAR, ErnieImage, and LongCatAudioDiTTransformer (that last one I'd missed in the first scan). Each gets AttentionMixin + a small regression test.

[github-actions]

Hi @HaozheZhang6, thanks for the PR! It does not appear to link an issue it fixes. If this PR addresses an existing issue, please add a closing keyword (e.g. Fixes #1234) to the PR description so the issue is linked. See the contribution guide for more details. If this PR intentionally does not fix a tracked issue, a maintainer can add the no-issue-needed label to silence this reminder.