Improve metrics processing

A few improvements:

- [ ] remove dependency on matplotlib from core fab. Maybe move the functionality into a dedicated command line tool?
- [ ] some phases do not appear to be tracked beyond bulk statistics, e.g. psyclone. Include the ability to track on a per-file basis.
- [ ] output data should be partitioned to split out summary statistics from individual data points
- [ ] fab command should include the ability to generate a text summary at the end of each run

Start by reading the existing metrics-processing path in core fab and the fab command entry point; the issue does not name specific files or tests. Map how matplotlib, phase statistics, output data, and run summaries are currently handled. Done means all four listed improvements have clear behavior and validation.

Written by the indexing model from the issue text.