What type of PR is this?

/kind feature
/kind api-change

What this PR does / why we need it:

This PR implements the DRA portion of KEP-4680, introducing a mechanism for the Kubelet to monitor and report the health of devices allocated via Dynamic Resource Allocation (DRA). This provides cluster operators and controllers with crucial visibility into device health directly in the Pod's status, enabling better troubleshooting and remediation of pod failures caused by faulty hardware.

The core changes are:

1. New gRPC Health Service (dra-health/v1alpha1):

   • Defines a new NodeHealth gRPC service with a WatchResources server-streaming RPC.
   • DRA plugins can optionally implement this service to stream device health status updates (Healthy, Unhealthy, Unknown) to the Kubelet.

2. Kubelet Integration:

   • The DRAPluginManager now detects plugins that implement the NodeHealth service and starts a long-lived WatchResources stream to monitor them. It gracefully handles plugins that do not implement the service.
   • The DRA Manager (pkg/kubelet/cm/dra) consumes these health updates and maintains a new, persistent healthInfoCache to track the last known state of each device. This cache handles device timeouts and survives Kubelet restarts.
   • When a device's health changes, the DRA Manager identifies all affected pods and triggers an asynchronous pod sync via the Container Manager.

3. API Change (v1.PodStatus):

   • A new field, allocatedResourcesStatus, is added to v1.ContainerStatus.
   • During a pod sync, the DRA Manager's UpdateAllocatedResourcesStatus function populates this field with the current health of the container's allocated devices, read from the health cache.
   • The status_manager's equality check (isPodStatusByKubeletEqual) is updated to correctly detect changes within this new field, ensuring the pod status is patched to the API server.

4. Testing:

   • Adds comprehensive unit tests for the new health caching and processing logic.
   • Adds a new node e2e test ([Feature:DynamicResourceAllocation] Resource Health) that verifies the end-to-end flow: a change in device health reported by a test driver is correctly reflected in the PodStatus on the API server.
   • Fixes discovered in existing e2e tests during development are included to ensure the suite remains stable.

Which issue(s) this PR fixes:

Fixes #126243

Special notes for your reviewer:

This PR focuses on the DRA implementation path for KEP-4680. All changes are behind the ResourceHealthStatus feature gate.

Does this PR introduce a user-facing change?

Yes. When the ResourceHealthStatus feature gate is enabled, a new allocatedResourcesStatus field will appear in the containerStatuses of a Pod's status if it uses Dynamic Resource Allocation. This allows users to inspect the health of the allocated devices.

The Kubelet can now monitor the health of devices allocated via Dynamic Resource Allocation (DRA) and report it in the `pod.status.containerStatuses.allocatedResourcesStatus` field. This requires the DRA plugin to implement the new v1alpha1 `NodeHealth` gRPC service. This feature is controlled by the `ResourceHealthStatus` feature gate.

Additional documentation e.g., KEPs (Kubernetes Enhancement Proposals), usage docs, etc.:

- [KEP]: https://github.com/kubernetes/enhancements/blob/master/keps/sig-node/4680-add-resource-health-to-pod-status/README.md