reform the structure for multi cloud

[yangw-dev]

What changed

Kubernetes base restructure: Split base/kubernetes/ into provider-agnostic base + provider overlays.

base/kubernetes/
├── base/                    # shared: harbor-namespace, nvidia-plugin, registry-mirror, git-cache
├── overlays/eks/            # EKS-specific: storageclass-gp3, node-performance-tuning
├── overlays/gke/            # GKE: shared base only (no provider resources yet)
└── kustomization.yaml       # top-level → overlays/eks (backward compat)

kubectl kustomize base/kubernetes/ output is identical before and after — zero diff.

Bootstrap multi-cloud routing: bootstrap-state.sh reads the cluster's cloud field and routes to bootstrap-state-gcp.sh for GCP clusters (GCS bucket, no DynamoDB needed). AWS path unchanged.

Path fixups: Updated justfile git-cache deploy path + doc references to match new structure.

Next steps

• deploy-module terraform init routing (S3 vs GCS backend)
• deploy / kubeconfig multi-cloud routing in justfile
• GKE terraform module (modules/gke/)
• GKE StorageClass
• Harbor values split (values.yaml + values-eks.yaml + values-gke.yaml)

[jeanschmidt]

I believe the changes are necessary, good that you took over this :)

as you are working on this, break the base module down so we can have a:

just deploy arc-cbr-production

and

just deploy arc-gcp-production

easily.

We can have the aws-specific stuff on one cluster and the gcp-specific stuff into another..

Maybe we should get rid of the base entierly, and have everything defined as modules that should be applied in sequence. I guess this would make this a more correct and flexible solutioin.