[Bugfix] fix mindiesd laserattention unsupported error

[fan2956]

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

Purpose

This PR fixes a RuntimeError encountered when calling the mindie custom op LaserAttention in diffusion models. The error occurred with the following message:
RuntimeError: call aclnnLaserAttention failed, detail:[PID: 1114958] 2026-04-10-10:18:07.438.311 AclNN_Parameter_Error(EZ1001): Get regInfo failed, The binary_info_config.json of socVersion [ascend910b] does not support opType [LaserAttention].
The root cause was that the ASCEND_CUSTOM_OPP_PATH environment variable was not properly configured before attempting to use the custom op. This PR ensures mindiesd is imported before the model's forward pass, which configures ASCEND_CUSTOM_OPP_PATH and allows LaserAttention to be called correctly.

Test Plan

export MINDIE_SD_FA_TYPE=ascend_laser_attention
vllm serve wan2.2-i2v-diffusers
--omni
--port 8099
--usp 8
--use-hsdp
--vae-patch-parallel-size 8
--vae-use-tiling
--log-stats
--profiler-config '{"profiler": "torch", "torch_profiler_dir": "./vllm_profile"}'

curl -X POST http://localhost:8099/v1/videos \
    -F "prompt='一只棕色野兔的正面特写镜头，采用低角度仰拍视角，营造亲密而庄严的视觉冲击。兔子一双圆润漆黑的大眼睛直视镜头深处，眼神中交织着野生动物的警觉与一丝难以言喻的温柔好奇，仿佛在与观者建立跨越物种的静默对话。它毛色呈现层次丰富的棕褐渐变，从浅奶油色腹部过渡到深棕背部，每根毛发纹理清晰可辨，在侧光下泛着丝绸般的光泽。细长洁白的胡须共有三对，随呼吸节奏微微颤动，偶尔因捕捉气流信息而轻轻摇摆。
两只标志性的长耳完全竖立，耳廓外侧覆盖短密棕毛，内侧则露出粉嫩的血管网络，薄如蝉翼的皮肤下血液流动隐约可见，耳朵以细微幅度不时转动，精准定位声源方向。背景是一片澄澈的蔚蓝天空，形态蓬松的白色积云以缓慢速度横向漂移，云影在兔子头顶交替变化，光线随之明暗流转。晴朗天气的明媚阳光从画面左上方45度角倾泻而下，在兔脸右侧形成柔和的伦勃朗式阴影，强化了面部立体感和皮毛质感。
兔子湿润的黑鼻子持续进行每秒三至四次的快速抽动，这是它们感知化学信号的本能动作，粉色三瓣嘴随之轻启，露出正在反刍的洁白门齿，下颌以稳定节奏左右研磨。摄影采用大光圈浅景深，焦点牢牢锁定在兔子双眼连线所在的焦平面，背景天空和远景绿色植被虚化成圆润的彩色光斑，前景几根嫩绿草叶闯入画面边缘，以缓慢弧线随风摇曳，暗示着和煦的春日微风。
整个场景弥漫着宁静致远的田园诗意，色彩温暖饱和，充满生命力。兔子在持续五秒的对视后，以典型 lagomorph 特征完成一次完整的瞬膜眨眼——第三眼睑从内侧横向滑过眼球，继而缓缓歪头向右十五度，这个行为在动物行为学中代表认知加工和好奇心表达，耳朵随之向同一方向倾斜，最终恢复正视姿态，胡须舒展，完成这段短暂而珍贵的自然纪录。'" \
    -F "input_reference=@/home/zf/vllm-omni/rabbit.jpeg" \
    -F "size=1280x720" \
    -F "seconds=5" \
    -F "fps=24" \
    -F "num_frames=121" \
    -F "guidance_scale=1" \
    -F "guidance_scale_2=1" \
    -F "flow_shift=5.0" \
    -F "num_inference_steps=4" \
    -F "seed=42"

Test Result

dit infer time
origin: 44s
Laserattention: 30s

---

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. Please provide the test scripts & test commands. Please state the reasons if your codes don't require additional test scripts. For test file guidelines, please check the test style doc
• [√] The test results. Please paste the results comparison before and after, or the e2e results.
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model. Please run mkdocs serve to sync the documentation editions to ./docs.
• (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 usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits.
Credits must be used to enable repository wide code reviews.

[gcanlin]

LGTM. Good catch!

[hsliuustc0106]

This is a duplicate of #2674 with the same diff. Please close this PR and keep #2674 (which has the ready label and passing CI).