Build autoscaler integration for taskbroker and workers

[markstory]

We're unlikely to get good resource utilization with CPU based autoscaling for workers, as it has not been overly effective for celery workers.

We'll need a way to have taskbrokers inform worker autoscaling. Ideally, as a broker instances's processing lag increases we scale up the worker pool. Conversely, when processing lag is low and task fetches start to miss we can scale worker pools down.

A potential solution is to build a prometheus style health/metrics endpoint into taskbroker that can be consumed by keda or k8s. We might be able to use the rabbitmq keda scaler integration as prior art as well.