FFI related paper cuts in driver ↔ lint-crate communication

[xFrednet]

When I created the prototype, I hacked together a basic driver ↔ lint-crate communication. After reworking the communication and learning more about FFI stuff, it turns out that the current implementation is unsound. This is a collection of all issues that need to be fixed:

• API types use non-#[repr(C)] types. These need to be replaced with FFI safe types or at least safely be transferred over the FFI boarder
  • Example: Slices, Option<>, &str
• Types from the FFI module should be hidden, as they are not part of the stable ABI
• Remove &dyn Trait pointers from the API
• Make sure that every exchanged type has #[repr(C)] (Remove traits from stable interface #6)
• Add CI with -Z randomize-layout #28
• Ideally: Lint to detect non #[repr(C)] type in #[repr(C)] struct rust-lang/rust-clippy#9357
• marker_api::LintPass::registered_lints returns Box<[..]> which is not FFI safe

Note: We can change #[repr(C)] to a different ABI as long as it's stable like C

[xFrednet]

I'm currently slowly working on these, while designing the AST. The last few PRs already made some good progress, like adding the FFI module to linter_api

[xFrednet]

A lot of progress on this has been made in #54 as it removes a lot of dyn and traits. The API is not totally FFI safe, but has a good baseline by my estimation