ffi-safe lint should be more aggressive in checking ffi stability of sub-types

[fschutt]

According to this answer, this code:

pub struct A { a: usize, b: usize }

#[repr(C)] 
pub struct B { 
    a: A 
}

... should emit a warning that struct A is still #[repr(Rust)], so the layout of B isn't completely defined, specifically the order of the fields a and b are not specified in #[repr(Rust)]. I'm not sure if the ffi-safe lint is active or not, but it would be very useful, so I know that if a type is marked #[repr(C)], all sub-types are also marked #[repr(C)].

Another question: When compiling to a cdylib, why does cargo use #[repr(Rust)] for publicly visible structs of a crate. It would be nice if, when compiling to a cdylib with this code:

pub fn foo(b: B) -> A { b.a }

... there could be automatic warnings:

• foo is a public function, should be marked as #[no_mangle]
• B and A should be marked as #[repr(C)]

... because the only representation that you usually want when compiling to a cdylib is #[repr(C)] (or #[repr(C, u8)] for enums).