Ideas for improving confidence in test case reduction

[afd]

• For all the Csmith programs that kill new mutants, find the union of all GCC and Clang warnings they give
• In the interestingness test, grep for warnings, and if any warning not in this set is exhibited, say "not interesting"
• This should avoid the need to check for specific warnings (but perhaps do so anyway as a safeguard?)

Also: run Frama-C against all fully reduced programs just in case it throws up anything else interesting.