Support the scale subresource on ModelDeployment

[negz]

Fixes #87

A ModelDeployment scales along one axis: spec.replicas. Each replica is a complete serving instance, so scaling means adding or removing whole replicas. Until now nothing exposed that axis through the Kubernetes scale subresource, so kubectl scale and event-driven autoscalers had no standard way to drive it. The design called for the subresource, but XRDs couldn't configure one until Crossplane v2.3 (crossplane/crossplane#7004), released recently.

This declares the scale subresource on the ModelDeployment XRD, mapping spec.replicas to the scale spec and status.replicas.total to the scale status. total is the count of scheduled replicas, which the composition function already writes; it mirrors how Deployment and LeaderWorkerSet report status.replicas as the observed total while keeping readiness in a separate field. The subresource declares no labelSelectorPath: a ModelDeployment's replica pods run on remote workload clusters, not alongside the XR, so a metric-based HorizontalPodAutoscaler can't observe them to scrape metrics. Scaling is by kubectl scale or an external-metrics autoscaler like KEDA. The project now requires Crossplane >=v2.3.0.

$ kubectl scale modeldeployment/qwen3-8b --replicas=3
modeldeployment.modelplane.ai/qwen3-8b scaled

Generating the Python schemas for an XRD with a scale subresource clobbered the ModelDeployment model with the autoscaling Scale type, which broke the composition functions that import it. crossplane/cli#119 fixes that in the schema generator; the crossplane-cli flake input (already pinned to main for an unreleased fix) moves forward to a main commit that includes it.

I have:

• Read and followed Modelplane's contribution process.
• Run nix flake check (or ./nix.sh flake check) and made sure it passes.
• Added or updated tests covering any composition function changes. No function changes; the XRD already had the status.replicas.total the subresource points at.
• Signed off every commit with git commit -s.

[Copilot]

Pull request overview

Adds Kubernetes scale-subresource support to the ModelDeployment XRD so standard tools (e.g., kubectl scale) and event-driven autoscalers can drive spec.replicas, and bumps the minimum Crossplane version to one that supports XRD scale subresources.

Changes:

• Declare the scale subresource on the ModelDeployment XRD, mapping spec.replicas and status.replicas.total.
• Require Crossplane >= v2.3.0 (XRD scale-subresource support) and update documentation accordingly.
• Advance the pinned crossplane/cli flake input to include the schema-generation fix for XRDs with scale subresources.

Reviewed changes

Copilot reviewed 4 out of 5 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
flake.nix	Documents why crossplane/cli remains pinned to main (now includes the scale-subresource schema-gen fix).
flake.lock	Updates the pinned crossplane/cli revision to a commit containing the needed generator fix.
design/design.md	Updates autoscaling design notes to reflect status.replicas.total as the scale status source.
crossplane-project.yaml	Bumps the minimum Crossplane version to >=v2.3.0 for XRD scale-subresource support.
apis/modeldeployments/definition.yaml	Adds the XRD subresources.scale mapping from spec.replicas to status.replicas.total.

---

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