[LTX-2] Add decode_audio to skip the audio decode in modular pipelines

[christopher5106]

What does this PR do?

Adds a per-call decode_audio input (default True) to the LTX-2 / LTX-2.5 modular pipelines. pipe(..., decode_audio=False) skips the audio VAE decode and vocoder and returns audio=None, for video-only outputs such as an IC-LoRA alpha matte (Lightricks/LTX-2.5-22b-IC-LoRA-Alpha-Gen). The video is unchanged, since the audio latents are still denoised jointly with it.

• New LTX2AutoAudioDecoderStep (ConditionalPipelineBlocks, modular_blocks_ltx2.py): runs LTX2AudioDecoderStep unless decode_audio=False, otherwise skipped. It is the same pattern as LTX2AutoPromptEnhancerStep.
• It replaces LTX2AudioDecoderStep in LTX2DecoderStep, LTX2ConditionDecoderStep, LTX25DecoderStep and LTX25ConditionDecoderStep, under the same audio_decode name: workflow layouts and the existing workflow tests are unchanged.
• LTX2AudioDecoderStep itself only gains the decode_audio input declaration, so the input shows up in the resolved workflows' inputs; it has no new branch.
• Tests: decode_audio=False gives audio=None and bit-identical videos, for LTX-2 text2video and LTX-2.5 in_context.
• Docstrings regenerated with utils/modular_auto_docstring.py, then doc-builder style --max_len 119.

Fixes #15044

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. [LTX-2] Modular pipelines: let callers skip the audio decode for video-only outputs #15044
• Did you make sure to update the documentation with your changes? (generated block docstrings; docs/source/en/api/pipelines/ltx2.md covers the non-modular pipelines only)
• 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)?

Self-review notes

Blocking: none.

Non-blocking, left for review:

1. Input declared but not read by the leaf. LTX2AudioDecoderStep declares decode_audio but does not read it (decoders.py). get_workflow() resolves conditional blocks down to their leaf, so declaring it there is what makes the input appear in a workflow's inputs. This mirrors enable_prompt_enhancement on the prompt-enhancer leaves. The alternative is overriding inputs on LTX2AutoAudioDecoderStep, but resolved workflows would then not list it.
2. Selection on a value. select_block keys on the value of a user input (decode_audio is False), not on its presence. The note in ConditionalPipelineBlocks.select_block restricts value-based selection for intermediate outputs only, and LTX2AutoPromptEnhancerStep does the same with enable_prompt_enhancement.
3. Generated docstrings change. The decoder chains now list audio_latents / audio_num_frames as optional: a conditional block without a default has no required inputs. LTX2AudioDecoderStep still requires them whenever it runs.

Dead code: none.
Verdict: READY.

Who can review?

@yiyixuxu @DN6