Fix Wan batch conditioning and config handling

[hlky]

Fixes #13578.

What does this PR do?

This PR addresses the Wan systemic review findings:

• Expands Wan i2v and Animate image embeddings, masks, latents, and conditioning tensors to batch_size * num_videos_per_prompt in per-prompt order.
• Keeps image required where VAE conditioning still needs it, while allowing precomputed CLIP image_embeds.
• Accepts documented Wan i2v image list/tuple inputs and validates precomputed image-embed batch sizes.
• Uses Wan VAE config scale factors in VACE, video-to-video, and modular Wan paths.
• Trims VACE reference latents for output_type="latent".
• Preserves Wan Animate image processor config values.
• Stops silently ignoring unsupported Wan video-to-video num_videos_per_prompt != 1.
• Keeps modular Wan timesteps in scheduler precision while casting model inputs to the selected transformer dtype.
• Adds focused fast regression tests and a .ai/pipelines.md checklist note for num_*_per_prompt expansion drift.

Tests

• ruff check ... on touched Python files
• ruff format --check ... on touched Python files
• git diff --check
• python -m compileall -q src/diffusers/pipelines/wan src/diffusers/modular_pipelines/wan tests/pipelines/wan tests/modular_pipelines/wan
• python -m pytest tests/pipelines/wan/test_wan_image_to_video.py tests/pipelines/wan/test_wan_vace.py tests/pipelines/wan/test_wan_video_to_video.py tests/pipelines/wan/test_wan_animate.py tests/modular_pipelines/wan/test_modular_pipeline_wan.py -q -k "not test_save_load_float16"

Result: 180 passed, 21 skipped, 5 deselected, 61 warnings, 6 subtests passed.

Note: WanFLFToVideoPipelineFastTests::test_save_load_float16 fails the same way on clean upstream/main (max diff 0.04175 > 0.01), so it was excluded from the clean touched-file verification.