Label GKE GPU node pools so the autoscaler can cold-start from zero

[negz]

Fixes #306.

A GKE GPU node pool with minNodeCount: 0 never scaled up. A model's pod claims its GPU through a DRA ResourceClaim, which binds only against a ResourceSlice the NVIDIA DRA driver publishes per running GPU node. At zero nodes there are no slices, and GKE's cluster autoscaler couldn't tell that a new node would satisfy the claim, so it refused to create the first one. The pod stayed pending with cannot allocate all claims.

This isn't a DRA architectural limitation. GKE's autoscaler recognises a DRA-capable node pool by the cloud.google.com/gke-nvidia-gpu-dra-driver=true node label and uses it to model the ResourceSlices such a node would publish when simulating a scale-up. compose-gke-cluster set only modelplane.ai/gpu and modelplane.ai/pool, so the pools were never recognised. This sets the label on every GPU node pool.

The getting-started GKE manifests pinned minNodeCount: 1 to work around the stall; they now allow minNodeCount: 0.

I kept the scope to the autoscaler label. GKE's DRA docs list three more node-pool requirements (gpu-driver-version=disabled, gke-no-default-nvidia-gpu-device-plugin=true, nvidia.com/gpu.present=true) for the manual-driver DRA path, but this repo runs the DEFAULT-driver path (GKE installs the driver at /home/kubernetes/bin/nvidia, layering the DRA driver on top) and #306 confirms DRA scheduling works once a node is up. The only gap for cold-start was the label.

Draft: not yet validated on a real GKE cluster. The scale-from-zero behaviour needs an end-to-end test against a GKE L4/A100 pool at minNodeCount: 0 before this merges.

docs/content/getting-started/what-youll-build.cast still shows the old minNodeCount: 1 line in its recording and needs regenerating separately.

[Copilot]

Pull request overview

This PR enables GKE GPU node pools to cold-start (scale up from minNodeCount: 0) for workloads that request GPUs via Kubernetes DRA ResourceClaims, by applying the node label that GKE’s cluster autoscaler uses to model DRA ResourceSlices during scale-up simulation.

Changes:

• Add the cloud.google.com/gke-nvidia-gpu-dra-driver=true node label to GPU node pools composed by compose-gke-cluster.
• Update compose-gke-cluster unit test expectations to include the new label.
• Update getting-started GKE manifests to allow minNodeCount: 0 for GPU pools (removing the prior workaround that pinned it to 1).

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated no comments.

File	Description
functions/compose-gke-cluster/function/fn.py	Adds a constant for the autoscaler-recognized label and applies it to GPU node pool labels.
functions/compose-gke-cluster/tests/test_fn.py	Updates golden expected node pool labels to include the new autoscaler label.
docs/manifests/getting-started/gke/platform.yaml	Changes GPU pool minNodeCount from 1 to 0 in the starter manifest.
docs/manifests/getting-started/gke/platform-scale.yaml	Changes GPU pool minNodeCount from 1 to 0 in the scale manifests.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.