fix: add /usr/lib to library search path in init container

[nmn3m]

Description

This PR fixes an issue where the kubelet-plugin init container fails to detect NVIDIA libraries installed in /usr/lib, causing the pod to remain stuck in Init:0/1 status indefinitely.

Fixes #692

After the fix:

$ kubectl get pods -n nvidia-dra-driver-gpu
NAME                                               READY   STATUS    RESTARTS   AGE
nvidia-dra-driver-gpu-controller-b65d7c4d9-rbr7t   1/1     Running   0          21m
nvidia-dra-driver-gpu-kubelet-plugin-vbjgc         2/2     Running   0          17s

Init container successfully completes:

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[nmn3m]

@klueska @jgehrcke , can you please take a look at this change?

[jgehrcke]

Maybe the patch is good, but we still need to figure out a strong reason for why we need this patch. Which guarantee is not fulfilled, and why? #692, as it stands, isn't quite revealing. For now, let's close this, and get back to this search path topic with more clarity later.

[nmn3m]

@jgehrcke @klueska — I had to write a custom bash script to manually copy the missing library to the path expected by the DRA driver:

echo "Copying NVIDIA libraries to worker node..."
docker exec kind-dra-gpu-worker mkdir -p /usr/lib/x86_64-linux-gnu
for lib in /usr/lib/libnvidia-ml.so*; do
  [ -e "$lib" ] && docker cp "$lib" kind-dra-gpu-worker:/usr/lib/x86_64-linux-gnu/
done

Is there a better way to handle this, or should we address it properly?