Fix Anima text encoder exports and standalone core denoising

[lucasruan1618]

Fix Anima text encoder exports and standalone core denoising

What does this PR do?

Fixes two issues when running Anima's modular stages as separate pipelines.

AnimaTextEncoderStep produces eight Qwen/T5 conditioning tensors, but none is tagged with kwargs_type="denoiser_input_fields". Consequently, the documented split-pipeline workflow:

text_embeddings = text_encoder_pipeline(prompt="a cat").get_by_kwargs("denoiser_input_fields")
output = pipeline(**text_embeddings, num_inference_steps=2)

exports an empty dictionary and fails with Required input 'qwen_prompt_embeds' is missing. This PR adds the tag to all eight outputs, applying the same fix pattern as #14925.

Separately, AnimaCoreDenoiseStep does not require a VAE, but latent preparation reads vae_scale_factor, whose property accesses self.vae directly. Running that block standalone therefore raises AttributeError: 'AnimaModularPipeline' object has no attribute 'vae_scale_factor'. This PR allows the property to return its existing default of 8 when the VAE component is absent. Pipelines with a VAE continue to use its configured scale factor.

The source changes are eight output tags and one component lookup. Regression tests cover split execution with omitted, empty, and non-empty negative prompts, comparing generated images against the complete pipeline. A separate test passes encoded tensors by name to the standalone core denoiser and checks the resulting latent shape and finite values.

Validation

• Anima modular suite: 57 passed, 5 skipped, including text-to-image and image-to-image tests.
• Shared TestBlockKwargsTypeInputs tests: 4 passed.
• make style, make fix-copies, make quality, and git diff --check: passed.
• Regression cases reproduced both original failures before the fixes.

Tests use real components from hf-internal-testing/tiny-anima-modular-pipe. Production-size checkpoints were not tested.

Self-review notes

Reviewed against .ai/references/review-rules.md, code_style.md, modular.md, and testing.md, following .ai/skills/self-review.

• Blocking issues: none.
• Non-blocking issues: none.
• Dead code: none introduced. Both changed paths are exercised through normal pipeline calls.
• Documentation impact: the existing split-pipeline example becomes usable for Anima; no API or usage-doc changes are needed.
• Verdict: READY. No unresolved findings.

Before submitting

• AI assistance was used for implementation and testing.
• Ran the self-review skill and included the final notes above.
• Added regression tests and ran formatting and consistency checks.
• Read the Coding with AI agents guide.
• Read the contributor guidelines and review this description before submitting.

Who can review?

@yiyixuxu

[github-actions]

Hi @lucasruan1618, 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.

Please note that PRs without a linked issue are likely to be automatically closed 10 days after this notice.

Once the PR links an issue (or gets the no-issue-needed label), you can ignore this message — it stays here as a comment, but it no longer applies.