升级到v2.9.0后，加载Eembedding和reranker模型报错 Could not load libtorchcodec.

### System Info / 系統信息 Name: torchcodec Version: 0.10.0 ### Running Xinference with Docker? / 是否使用 Docker 运行 Xinfernece？ - [x] docker / docker - [ ] pip install / 通过 pip install 安装 - [ ] installation from source / 从源码安装 ### Version info / 版本信息 v2.9.0 ### The command used to start Xinference / 用以启动 xinference 的命令 docker run -d \ --name xinference2 \ --gpus '"device=1"' \ -p 9995:9997 \ --ipc=host \ --shm-size 64G \ --restart unless-stopped \ -e VLLM_WORKER_MULTIPROC_METHOD=spawn \ -e VLLM_LOGGING_LEVEL=INFO \ -e XINFERENCE_LOG_LEVEL=INFO \ -e HF_HOME=/root/.cache/huggingface \ -e MODELSCOPE_CACHE=/root/.cache/modelscope \ -v /mnt/raid_storage/share/xinference-shared/huggingface:/root/.cache/huggingface \ -v /mnt/raid_storage/share/xinference-shared/modelscope:/root/.cache/modelscope \ -v /mnt/raid_storage/share/xinference-shared/xinference-huggingface:/root/.xinference/huggingface \ -v /opt/xinference2/cache:/root/.xinference/cache \ -v /opt/xinference2/virtualenv:/root/.xinference/virtualenv \ -v /opt/xinference2/model:/root/.xinference/model \ -v /opt/xinference2/openmind_hub:/root/.xinference/openmind_hub \ -v /opt/xinference2/flashinfer:/root/.cache/flashinfer \ xprobe/xinference:v2.9.0 \ xinference-local -H 0.0.0.0 --metrics-exporter-host 0.0.0.0 --metrics-exporter-port 9401 ### Reproduction / 复现过程 升级到v2.9.0后，加载Embedding和reranker模型都不成功，会报错。与之前修复过的问题一样https://github.com/xorbitsai/inference/pull/4817，回退到v2.7.0后可以正常加载。 ### Expected behavior / 期待表现 希望可以正常加载