Configure Karpenter NodePool resource limits to cap total CPU and memory provisioned

Steps

1. Add a spec.limits block to the NodePool with `cpu` and `memory` fields (e.g. `cpu: "1000"`, `memory: 1000Gi`) to cap the total sum of resources Karpenter will provision under that specific NodePool.
2. Use a DecimalSI string value for cpu (avoid bare unquoted integers, since the API coerces them to strings and can cause GitOps drift/diff skew) and a BinarySI value like `1000Gi` for memory.
3. Optionally cap extended resources such as `nvidia.com/gpu` under the same spec.limits block, and use `limits.nodes` to cap node count directly.
4. Apply spec.limits per NodePool if different teams or workloads need separate caps, since limits are scoped to a single NodePool rather than being a cluster-wide setting.
5. Monitor current consumption against the configured limits with `kubectl get nodepool -o=jsonpath='{.items[0].status}'` or by inspecting the NodePool's status.resources field.
6. When a limit is reached, Karpenter blocks further node provisioning under that NodePool until existing nodes are terminated or consolidated to free headroom — it does not evict running pods to enforce the cap.