Add fuzzing to the CI

[MarcelGarus]

Depends on #245

This makes the exit code of the candy fuzz command match whether errors were found during fuzzing. It also adds fuzzing to the CI and fixes some errors that were discovered.

[MarcelGarus]

In the logs, you may have noticed that the functions in Core are fuzzed several times. That's because the tree of modules created from Core/_.candy contains the outer leaf files many times. For example, the core.bool module is imported by many other modules in the core package, causing it to be instantiated and evaluated many times. All the created closures are separate and fuzzed independently.

I guess I'll start working on some basic optimizations such as function inlining, duplicate subtree elimination, and compile-time evaluation soon to fix the problem described above. LLVM would also do those optimizations, but we'd have to turn the entire tree (including expanded uses, thus an exponential amount of code) into LLVM-IR for it to do its magic, so that's too late in the pipeline and it makes sense to do these optimizations in Candy itself.

[JonasWanke]

Rust also has a few complaints:

[MarcelGarus]

Fuzzing works fine locally. Seems like this is an issue of nightly cargo: rust-lang/cargo#8557

[JonasWanke]

I'm not sure whether the issue you linked is really related since it's been closed for more than two years, but I couldn't find a good-looking current one. Maybe we should disable the action for now so that we don't get used to the failing CI?

Other than that, the PR looks fine

[MarcelGarus]

Fair enough, I'll just disable the fuzzing for now