Private AI service container: GPU support enhancements please

Currently, when creating a new PrivateAI resource in OraOperator 2.2.0,
one can only specify the worker node name where the PAI container is supposed to run on.
Especially in GPU enabled environments, I am requesting three enhancements:

1. instead of specifying only a workernode host name, please let the user set his own nodeSelector clause. In my OKE environment, GPU enabled nodes get their own labels to select from. So allowing this selector syntax would be great:

        nodeSelector:
          node.kubernetes.io/instance-type: VM.GPU.A10.1

2. GPU enabled nodes often have taints attached to them. To be able to run on that kind of node, the PAI container must tolerate the tainted node. In my environment, this looks like the following:

        tolerations:
        - effect: NoSchedule
          key: nvidia.com/gpu
          operator: Exists

3. The GPU-enabled PAI container will not start if we dont allow him to use or see GPU devices. please set this environment variable in PAI deployments or allow specifying which GPU devices should be reachable.

          - name: NVIDIA_VISIBLE_DEVICES
            value: ALL

Start by locating the PrivateAI resource definition and the deployment logic that creates its PAI container. Check how Kubernetes node selection, taints, tolerations, and container environment variables are represented. Done means users can configure nodeSelector and tolerations and can expose the requested GPU devices for PrivateAI deployments, with corresponding tests.