More extensive, systematic comparison of HarfBuzz and HarfRust shaping

Although HarfRust (and RustyBuzz originally) is a direct port from C to Rust, there might still be undiscovered shaping differences due to the porting process.

To increase confidence that HarfRust matches exactly HarfBuzz shaping (where it is applicable), I suggest we do extensive shaping tests of large corpus of texts and a large corpus of fonts (e.g. Google Fonts collection, Noto fonts, SIL fonts, etc.)

I imagine a test shaping harness that can take a collection of texts and fonts and shape them, compare the outputs, and report any differences. It could also try create a minimal test case for each failure to be added to HarfBuzz and HarfRust test suites.