We don't imply outlives-bounds from the own explicit bounds of a GAT projection

[fmease]

Not sure if worth tracking/discussing. I'm basically revisiting this because of IATs and LTAs. I know that T-types would like to avoid implying outlives-bounds in more places and under more circumstances at least until various soundness issues related to them have been fixed (LTAs prolly being the exception) (via: #t-types > implied bounds for the more recent features? @ 💬 (2023)).

Presently, when inferring outlives-bounds we only consider the explicit predicates of the corresp. trait when looking at projections as noted by this comment (which was written by me):

rust/compiler/rustc_hir_analysis/src/outlives/implicit_infer.rs

Lines 207 to 209 in 77101fe

	// This corresponds to a type like `<() as Trait<'a, T>>::Type`.
	// We only use the explicit predicates of the trait but
	// not the ones of the associated type itself.

This means:

trait Trait<'a, T: 'a> { type Project; }
impl<'a, T: 'a> Trait<'a, T> for () { type Project = (); }

struct Env<'a, T>(<() as Trait<'a, T>>::Project);
//                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ from this projection, we imply `T: 'a`

//fn env<'a, T>() { let _: Env<'a, T>; } //~ ERROR the parameter type `T` may not live long enough

trait Trait { type Project<'a, T: 'a>; }
impl Trait for () { type Project<'a, T: 'a> = (); }

struct Env<'a, T>(<() as Trait>::Project<'a, T>); //~ ERROR the parameter type `T` may not live long enough
//                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ we **don't** imply anything from this projection!

[fmease]

Similarly ofc here:

struct Env<'a, T: Trait<'a>>(<T as Trait<'a>>::T); //~ ERROR the parameter type `T` may not live long enough
trait Trait<'a> { type T where Self: 'a; }