Better support for understanding performance

[bernhold]

Initiated by Mark Cianciosa on 2021-10-08 ORNL-AToM call.

• Let's make it easy to get trace information out of IPS so users can understand component runtimes, look for bottlenecks, etc.
• Let's provide simple job-end summaries, such as min/avg/max/sd of run time for each component, etc.
• Let's make it possible to understand resource utilization throughout a job
• Let's provide simple job-end summaries of resource utilization

[markcoletti]

Related to this, there needs to be more logging added to inform the users about what's going on. I've noticed in struggling to get the tutorials to run that things ... just fail. Sure, there's a stack trace, but it would've been helpful if --verbose and even --debug toggled additional logging to inform me where in the initialization process things are failing. Because now I'm going to have to step through ips.py with a debugger to understand where things are going off the rails.