Supplement generated tests with a fuzzing framework

[alice-i-cecile]

trafficstars

1. The test generating code is broken, and hard to maintain, see #89 and #65.
2. The automatic tests are massive, resulting in painful renames.
3. This strategy is extremely hard to apply to alternative layout strategies (#28).
4. These generated tests aren't even complete, and have failed to catch multiple deviations from the spec.

These should be supplemented with a fuzz testing or property-based testing framework.

[alice-i-cecile]

Checking for Chrome compliance was a solid idea though, and should be retained in CI.

[mockersf]

I re-added (and fixed) the gentest script in #145. I believe it can't be replaced by fuzzing or prop testing, as both those techniques would need a third party flexbox implementation to generate the assertions anyway.