[Feature] Support cache-dit for Wan 2.2 inference

[SamitHuang]

PLEASE FILL IN THE PR DESCRIPTION HERE ENSURING ALL CHECKLIST ITEMS (AT THE BOTTOM) HAVE BEEN CONSIDERED.

Purpose

Accelerate Wan2.2 series generation.

• Fix the bug of dummy run impact on diffusion cache refresh
• Fix and tune cache-dit config for higher performance of the Wan2.2 pipelines

Test Plan

python text_to_video.py \
  --prompt "Two anthropomorphic cats in comfy boxing gear and bright gloves fight intensely on a spotlighted stage." \
  --negative_prompt "色调艳丽，过曝，静态，细节模糊不清，字幕，风格，作品，画作，画面，静止，整体发灰，最差质量，低质量，JPEG压缩残留，丑陋的，残缺的，多余的手指，画得不好的手部，画得不
好的脸部，畸形的，毁容的，形态畸形的肢体，手指融合，静止不动的画面，杂乱的背景，三条腿，背景人很多，倒着走" \
  --height 480 \
  --width 832 \
  --num_frames 33 \
  --guidance_scale 4.0 \
  --guidance_scale_high 4.0 \
  --boundary_ratio  0.875 \
  --num_inference_steps 40 \
  --fps 16 \
  --cache_backend cache_dit \
  --enable-cache-dit-summary \
  --output t2v_out.mp4

Test Result

'CFG Cache Steps: 15, [2, 3, 4, 6, 7, 8, 10, 11, 12, 14, 15, 16, 18, 19, 21]'
("CFG Residual Diffs: 17, {'2': 0.03515625, '3': 0.06494140625, '4': "
 "0.0966796875, '6': 0.0390625, '7': 0.0732421875, '8': 0.11083984375, '10': "
 "0.047607421875, '11': 0.091796875, '12': 0.140625, '14': 0.060546875, '15': "
 "0.11962890625, '16': 0.185546875, '18': 0.08984375, '19': 0.185546875, '20': "
 "0.306640625, '21': 0.1796875, '22': 0.44140625}")
[Stage-0] INFO 01-28 02:21:52 [diffusion_engine.py:74] Generation completed successfully.
[Stage-0] INFO 01-28 02:21:52 [diffusion_engine.py:97] Post-processing completed in 0.1659 seconds
INFO 01-28 02:21:53 [log_utils.py:550] {'type': 'request_level_metrics',
INFO 01-28 02:21:53 [log_utils.py:550]  'request_id': '0_fe657b0b-4321-40d4-893c-1c6d71568240',
INFO 01-28 02:21:53 [log_utils.py:550]  'e2e_time_ms': 33146.83222770691,
INFO 01-28 02:21:53 [log_utils.py:550]  'e2e_tpt': 0.0,
INFO 01-28 02:21:53 [log_utils.py:550]  'e2e_total_tokens': 0,
INFO 01-28 02:21:53 [log_utils.py:550]  'transfers_total_time_ms': 0.0,
INFO 01-28 02:21:53 [log_utils.py:550]  'transfers_total_bytes': 0,
INFO 01-28 02:21:53 [log_utils.py:550]  'stages': {0: {'stage_gen_time_ms': 32580.10959625244,
INFO 01-28 02:21:53 [log_utils.py:550]                 'num_tokens_out': 0,
INFO 01-28 02:21:53 [log_utils.py:550]                 'num_tokens_in': 0}}}
Processed prompts: 100%|██████████████████████████████████████████████████████████████████████████████| 1/1 [00:33<00:00, 33.15s/img, est. speed stage-0 img/s: 0.00, avg e2e_lat: 0.0ms]
INFO 01-28 02:21:53 [omni.py:833] [Summary] {'e2e_requests': 1,███████████████████████████████████████| 1/1 [00:33<00:00, 33.15s/img, est. speed stage-0 img/s: 0.00, avg e2e_lat: 0.0ms]
INFO 01-28 02:21:53 [omni.py:833]  'e2e_total_time_ms': 33148.048400878906,
INFO 01-28 02:21:53 [omni.py:833]  'e2e_sum_time_ms': 33146.83222770691,

t2v_out.cachedit.mp4

E2E time is reduced from 81.55s to 33s, 2.47x acceleration.

---

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan, such as providing test command.
• The test results, such as pasting the results comparison before and after, or e2e results
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.
• (Optional) Release notes update. If your change is user facing, please update the release notes draft.

BEFORE SUBMITTING, PLEASE READ https://github.com/vllm-project/vllm-omni/blob/main/CONTRIBUTING.md (anything written below this line will be removed by GitHub Actions)

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: b19b0805c1

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[david6666666]

LGTM