agent should surface metrics on job completion times

[psFried]

A frequent pain point is the time it takes the agent to complete async jobs. It executes jobs sequentially, which means that head of line blocking is an issue. We like to surface some better metrics around this. Some metrics that seem useful are:

• Time each job spends in queue before being started
• Time each job takes to complete, once started
• Queue depth, if it's practical to get that

[jgraettinger]

We do have nearly all of this available in the tables already. Queue depth, for example, is a count of rows with status "queued". Total execution time is the difference between the creation & last-update timestamps.