add torchvision: Qwen3-VL processor imports it at load time

Qwen3-VL-Embedding-2B server on ZeroGPU

initial commit