There are some areas of the regexp module that are not covered by the tests, as extensive as they already are.

Mostly, they include some prefix stuff (that may have caused #30511 & #30425), some utf8 code paths, some case folding, some paths in onepass and the RuneReader input interface.

Does the idea of trying to run all test regexps through all 3 matchers (onepass when possible, backtrack, nfa) and all 3 input types (string, bytes, reader) sound good? We should aim for each of them being individually correct, and this would help adding new ones in the future (dfa!).

I'm starting work on a CL. I want to throw in some of the fuzzing I did before in #21463, should I add it here or open a separate issue/CL?

Thanks!