A platform team runs its self-hosted Actions infrastructure on Kubernetes and wants a supported foundation for runner scale sets and autoscaling. Which project is the appropriate starting point?

1. Use a generic Kubernetes pod autoscaler with no Actions controller and assume it handles runner registration and job assignment.
2. Use only a workflow concurrency group as the infrastructure autoscaler.
3. Actions Runner Controller (ARC).
4. A workflow template containing a larger job matrix.