[batch] Use a simulated job queue to estimate the ready cores in the control loop

[jigold]

I added a new Grafana panel without alerts that hopefully will let us catch problems if jobs aren't getting scheduled in a timely manner. I think to have an alert, we'd want to measure what the average wait time of a job in the queue is which would require more infrastructure (keeping track of last state change). We can consider adding that now -- not sure how much work it would be.

[jigold]

I changed the control loop to use the old method of estimating ready cores for now. I'll change it to the new way in a second PR. This way we can make sure we're happy with the metrics in Grafana before I change the behavior.

[jigold]

This is ready for a look. The tests are passing except one service backend test timed out.