Improve test coverage and create infrastructure for integration tests

[dubiousconst282]

We could also reuse some of the runtime/JIT tests for "correctness"/integration tests:

• https://github.com/dotnet/runtime/blob/main/docs/project/jit-testing.md
• https://github.com/dotnet/runtime/tree/main/src/tests/JIT

It's clearly impossible to catch edge cases with only traditional testing. Fuzzing seems to be a popular approach to tackle this:

• https://github.com/jakobbotsch/Fuzzlyn
• https://cfallin.org/blog/2021/03/15/cranelift-isel-3/