Present profiles nicely

Several bits:

- [ ] At the moment, a separate profile appears for each thread. Instead, only present a single profile with thread max/min info.
- [ ] Write to a suitable output file.
- [ ] Sort the profile entries in order of decreasing self time; perhaps optionally the total time (inclusive of child routine calls).

Start by locating Vernier’s code that assembles and presents profile entries, then trace how profiles are currently written to output. Done means each thread contributes to one profile with min/max information, results are written to a suitable output file, and entries can be ordered by decreasing self time with optional inclusive-time ordering.

Written by the indexing model from the issue text.