Add DRA support for GPU pod eviction during driver upgrades

[karthikvetrivel]

The driver-manager init container evicts GPU pods from a node before reloading the NVIDIA GPU driver, but today it only detects pods that request GPUs via device-plugin resources (nvidia.com/gpu, nvidia.com/mig-*). This PR extends the pod filter to also detect pods using GPUs through Dynamic Resource Allocation (DRA): for each pod on the node, it resolves the referenced ResourceClaim(s) and checks whether any is allocated by the NVIDIA GPU DRA driver (gpu.nvidia.com).

Testing

I tested the following scenarios on a two node k8s 1.34 cluster with two GPUs and the NVIDIA DRA driver installed:

1. Create a pod that uses a GPU via a directly-named DRA ResourceClaim (allocated by gpu.nvidia.com) alongside a non-GPU pod on the same node. Verify the claim is allocated and reserved for the pod.
2. Trigger the driver-manager eviction flow (uninstall_driver). Verify the DRA GPU pod is identified and evicted, while the non-GPU pod is skipped.
3. Verify detection against the live ResourceClaim API across claim shapes: directly-named and templated claims allocated by gpu.nvidia.com are detected, while compute-domain.nvidia.com, other-driver, unallocated, and not-found claims are not.
4. Verify traditional device-plugin pods are still detected via nvidia.com/gpu and nvidia.com/mig-* resource requests.