Code review comment for lp:~vorlon/lp-ftbfs-report/report-only-regressions

New commit speeds up a full report run to where we pay only a ~2.5x runtime penalty for the added processing, instead of ~3.3x. I don't see an obvious way to eke out much more.

Saw some failures incorrectly go missing from the report earlier in testing which I can now no longer reproduce. But this may be worth spot-checking a few more runs before landing.

Presently, the amended results are 40->38 packages for main (only firefox and thunderbird now dropped out), but 1424->1152 for universe.