Report change in RPITIT lifetime capture clauses.

Forbid RPITIT from implementations from capturing more lifetimes than the trait definitions allows.

…ures, r=oli-obk

Error when RPITITs' hidden types capture more lifetimes than their trait definitions

This implements a stricter set of captures rules for RPITITs. They now may only capture:
1. Lifetimes from the impl header (both the self type and any trait substs -- we may want to restrict just to the self type's lifetimes, but the PR makes that easy to do, too)
2. Lifetimes mentioned by the `impl Trait` in the trait method's definition.

Namely, they may not mention lifetimes from the method (early or late) that are not mentioned in the `impl Trait`.

cc rust-lang#105258 which I think was trying to do this too, though I'm not super familiar with what exactly differs from that or why that one was broken.
cc rust-lang#112194 (doesn't fix this issue per se, because it's still an open question, but I think this is objectively better than the status quo, and gets us closer to resolving that issue.)

Technically is a fix for the ICE in rust-lang#108580, but it turns that issue into an error now. We can decide separately whether or not nested RPITITs should capture lifetimes from their parents.

r? `@oli-obk`

…ures, r=oli-obk

Error when RPITITs' hidden types capture more lifetimes than their trait definitions

This implements a stricter set of captures rules for RPITITs. They now may only capture:
1. Lifetimes from the impl header (both the self type and any trait substs -- we may want to restrict just to the self type's lifetimes, but the PR makes that easy to do, too)
2. Lifetimes mentioned by the `impl Trait` in the trait method's definition.

Namely, they may not mention lifetimes from the method (early or late) that are not mentioned in the `impl Trait`.

cc rust-lang#105258 which I think was trying to do this too, though I'm not super familiar with what exactly differs from that or why that one was broken.
cc rust-lang#112194 (doesn't fix this issue per se, because it's still an open question, but I think this is objectively better than the status quo, and gets us closer to resolving that issue.)

Technically is a fix for the ICE in rust-lang#108580, but it turns that issue into an error now. We can decide separately whether or not nested RPITITs should capture lifetimes from their parents.

r? ``@oli-obk``