feat(gke): add gpuSharing field to ComputeClass GPU spec

[davidbellemare]

1. Please confirm that you have read the document before PR submitted

• Yes, I have read THIS NOTE DOC.

2. Contact Information(Optional)

If it is convenient, please provide your contact information so we can reach you when processing the PR:

• Email: bellemare.david@gmail.com
• HomePage:

---

What does this PR do?

Adds gpuSharing support to the spec.priorities[].gpu field of the GKE ComputeClass CRD schema.

GKE 1.35.2-gke.1485000 introduced a gpuSharing object under the GPU section that enables:

• Time-sharing / MPS — multiple containers share a single physical GPU
• MIG partitioning — the GPU is split into isolated multi-instance partitions (gpuPartitionSize)

Without this change, any ComputeClass that uses gpuSharing fails KCL schema validation with an unknown-attribute error.

New schema

A new CloudGoogleComV1ComputeClassSpecPrioritiesItems0GpuSharing schema is added:

Field	Type	Description
sharingStrategy	str	TIME_SHARING or MPS
maxSharedClientsPerGPU	int	2–48; required with a sharing strategy
gpuPartitionSize	str	MIG profile, e.g. "1g.24gb". Omit to skip partitioning.

Example

priorities = [
    {
        gpu = {
            count = 1
            driverVersion = "latest"
            type = "nvidia-rtx-pro-6000"
            gpuSharing = {
                gpuPartitionSize = "1g.24gb"
            }
        }
        machineType = "g4-standard-48"
    }
]

Ref: https://cloud.google.com/kubernetes-engine/docs/reference/crds/computeclass