Qwen3omni ViT Runner

class Qwen3OmniViTRunner : public trt_edgellm::rt::Qwen3VLViTRunner

Qwen3-Omni vision encoder. Mirror of HF modeling_qwen3_omni.py, whose ViT IS Qwen3-VL’s (fast-pos-emb + deepstack — inherited unchanged) but whose preprocessing/MRoPE follow Qwen2.5-VL: a flat video span, 2D resize, constant pad fill, position_id_per_seconds, and an MRoPE advance that also counts the T-extent.