fix(vllm): make vllm-omni respect the selected torch backend

[rushilbhat]

Summary

Fix a Docker build failure in the vLLM source-install path by making the vllm-omni install respect the selected Torch backend.

Problem

When install_vllm.sh is used to install vLLM from source in a CUDA environment, it installs vllm-omni before building vLLM.

The issue was that the vllm-omni install did not use the selected Torch backend. That could pull in a Torch build for a different CUDA version than the one expected by the image being built, which caused the Docker build to fail with a CUDA/Torch version mismatch.

Solution

Update container/deps/vllm/install_vllm.sh so that CUDA installs of vllm-omni use the same --torch-backend selected for the rest of the vLLM install flow. This applies to both the PyPI install path and the source fallback path