Handle GPU partitioning mode changes on the same Node (MIG<>MPS)

[Telemaco019]

Problem description

When changing the partitioning mode of a node from MPS to MIG, the nvidia-device-plugin crashes and therefore any new MIG device created by nos is never exposed to k8s as resource.

How to reproduce

1. Enable MPS partitioning on a node (kubectl label nodes <node> "nos.nebuly.com/gpu-partitioning=mps")
2. Create a Pod requesting MPS resources (for instance nvidia.com/gpu-10gb)
3. After the requested MPS resources are created and the Pod is scheduled on the node, delete the Pod and change the node's GPU partitioning mode to MIG (kubectl label nodes <node> "nos.nebuly.com/gpu-partitioning=mps")
4. Create a Pod requesting MIG resources (for instance nvidia.com/mig-1g.10gb)

Expected behaviour

After step 4, the MIG resources are created automatically and the Pod is scheduled on the node

Actual behaviour

After step 4, the MIG devices are created on the GPU, however the nvidia-device-plugin Pod crashes with error Cannot find configuration named <config-name>, where <config-name> is the name of the configuration set by nos during step 2.