HPA: Scale up fast, but scale down slooowly

[henrik242]

We're having issues with the default Horizontal Pod Autoscaler scaling up and down aggressively (and in large chunks), going from 20 to 14 to 20 to 11 to 20 (etc) in a relatively short period of time.

Setting replicas.cpu_threshold_percentage: 70 has helped a little bit, but not much, as the app is seeing bursts of > 160% cpu from time to time.

Ideally we'd want to scale up fast, but scale down slowly. FIAAS should expose the knobs for configuring this, or provide an alternative nicely behaving HPA.

[henrik242]

Depends on #124