trait_added_supertrait should check for blanket impls and their bounds

[obi1kenobi]

The trait_added_supertrait lint (#441, #892) is unable to do precise-enough analysis of supertraits (including foreign traits, blocked on #638) since generics and bounds aren't modeled in our schema yet (#149).

This can cause false-positives in (arguably, relatively pathological) cases such as:

// old:
pub trait Example {}

// new:
pub trait Super {}

impl<T> Super for T {}

pub trait Example: Super {}

Transitive blanket impls can also cause false-positives (an even more-pathological case):

// old:
pub trait Example {}

// new:
pub trait MoreSuper {}

impl<T> MoreSuper for T {}

pub trait Super {}

impl<T: MoreSuper> Super for T {}

pub trait Example: Super {}

In both of these cases, the added supertrait bound is trivially satisfied due to blanket impls, so it's not a major breaking change to add it.

Resolving this issue requires checking (transitively!) for universal blanket impls of the specific ImplementedTrait (trait + generics) that was added in the bound.

It has to be ImplementedTrait ("trait + generics") and not just Trait because it's possible the blanket is only for that specific parameterization of the trait.

For example: impl<T> Blanket<'static, ()> for T {}.

Blankets that don't fit the added supertrait bound don't apply, and in such cases the change is still breaking.

For example: impl<T, U: Copy> Blanket<U> for T {} will not prevent a breaking change if a trait adds a bound like

• trait Example<T>: Blanket<T>, where T: Copy isn't part of the trait's bounds, or like
• trait Example: Blanket<String> where String isn't Copy.