Set their resource requests low, (but try and ensure we don't hit OOMs
as a result)

And limit their maxReplicas to 1 more than their minReplicas. (This way
we are still testing the pod autoscaling, but we aren't eating review
resources)