[dynamic infra] Correctly manage Save Code Now workers autoscaling

Kubernetes is not really smart to downscale running pods, it choses randomly a pod to kill.

An usual scenario is:

• 2 save code now messages arrive in the rabbitmq queue
• task 1 is a long task
• task 2 is a short task
• 2 pods are created
• task 2 is done
• kubernetes choses a random pod to kill
• A warm shutdown message is sent to pod1
• If the task is not finished after 1 hour, the pod is killed
• Pod2 is doing nothing in the interval

Possible solutions:

• Use scaled job instead of scaled object: the loaders must be able to load only one repo and exist
• Find a way to manage the controller.kubernetes.io/pod-deletion-cost of the pods doing nothing
• ???

The easy workaround is to disable the autoscaling of the high priority loaders until a solution is found.