effective visibilities: Split reachable into "reachable through interfaces" and "reachable through bounds"

[petrochenkov]

We currently have two lints as a part of #48054 - private_interfaces and private_bounds, the former being the stronger one.

struct A {}
struct B {}

pub fn leak() -> A // Warning: private_interfaces
    where B: Sized // Warning: private_bounds
{
    A {}
}

The detailed distinction is documented in the RFC, but the general idea is that with private_interfaces we can get our hands on a value of the private type A (which will result in a hard error), and with private_bounds we cannot.

In the effective visibility table it is likely reasonable to make exactly the same distinction - items reachable through interfaces (higher level of reachability), and items that are only reachable through bounds (less reachable).

In particular, the unnameable_types lint will only want to report types reachable through private interfaces, to reduce false positives.

This is a follow up to #48054.