Track latency and queue duration SLOs for each worker individually

In our first iteration of tracking worker durations we relied on the priority label.

It turns out this this is not a good way to compare worker durations as there is little logic to distinguish the SLOS of say besteffort workers from realtime workers: some realtime jobs can take 10 minutes to run, some besteffort jobs, milliseconds.

For this reason, it makes more sense to track an apdex score individually for each queue.