kube: controller lifecycle improvements

[milas]

• Optimize the Dockerfile to use cache mounts
• Use the context.Context during HTTP operations
• Add an internal context for normal controller operations
• Sequence shutdown to wait for informer to stop and then attempt to unexpose any ports before exiting
  • Unexpose ports in parallel, we just log out all the errors regardless