[skrifa] assess autohinting performance

Testing with fauntlet revealed that applying autohinting to the [Handjet](https://fonts.google.com/specimen/Handjet) font takes considerable time to the point where we should likely avoid doing it at all.

Let's get some metrics for various fonts (particularly on mobile) and try to come up with reasonable limits to avoid performance cliffs.

Note: cases where autohinting takes excessive time likely indicate very complex or overly dense geometry which will either not benefit from hinting or produce distorted results.

Start by reproducing the Handjet autohinting case with fauntlet, then measure autohinting time across various fonts, particularly on mobile. Compare the results to geometry complexity and define reasonable limits that avoid performance cliffs; done means the limits and their expected behavior are established and covered by the relevant measurements.

Written by the indexing model from the issue text.