DREIMP-10951: Autoscale VitessCell pods with HPA

[siadat]

Problem

We'd like to autoscale the vtgate deployments created by the vitess-operator for each VitessCell.

Solution

Create an HPA for each VitessCell, where max_instances is set.

This PR should be merged after #3940

Context

• The VitessCell CRD should be changed as well: https://github.yelpcorp.com/misc/compute-infra-k8s/pull/1949
• Example autoscaled instance: https://github.yelpcorp.com/sysgit/yelpsoa-configs/pull/48782

http://y/DREIMP-10951

[nemacysts]

imo, the autoscaling for vitess seems different enough that it's probably more maintainable to have a separate script that we use to configure HPAs rather than mix it into the existing code for long-running services

[siadat]

imo, the autoscaling for vitess seems different enough that it's probably more maintainable to have a separate script that we use to configure HPAs rather than mix it into the existing code for long-running services

@nemacysts I added this method override to avoid using the existing one in KubernetesDeploymentConfig.

Afaiu, atm the HPA for the deployments created by VitessCell look like a regular HPA that uses CPU metrics... but I guess we'd like to separate autoscaling deployments created by CRs vs otherwise in general and in case they diverge in the future? so I copied the code from KubernetesDeploymentConfig to this class.