Handling of #[test] is a hack

[alastairreid]

The right way to support #[test] in a formal verification tool is almost certainly to implement it as part of rustc because that is how it is supported for testing.
Instead, we jump through a number of hoops to

• make a list of all the tests
• find the mangled names of tests
• invoke each test individually
• find whether a test #[should_panic]
• run N copies of the verifier in parallel

This allows us to write tests that work for both dynamic verification (testing / fuzzing) and static verification (symbolic execution, model checking, ...) but the result is that scripts/cargo-verify is a collection of hacks to do the above.

The current approach makes it possible to experiment with the interface and understand the issues - but, in the long run, some changes to rustc will probably be required.