implied-bounds and normalization

[lcnr]

Affected tests:

• implied bounds for opaques now normalizes opaques, unsound (normalizing opaques while computing implied bounds #159)
  • tests/ui/impl-trait/unactionable_diagnostic.rs
• now compiles? cc incomplete normalization in implied bounds rust#109799
  • old solver has the nested Projection(<&'x str as IntoIterator>::Item, &'?x str) which gets ignored while new solver has <Vec<&'x str> as Iter>::Item: 'static which gets normalized to &'x str in one step, proving the Projection(<&'x str as IntoIterator>::Item, &'?x str) goal inside of the trait solver
  • tests/ui/implied-bounds/normalization-nested.rs#lifetime
• computing implied bounds does not normalize correctly (fixed by Properly deeply normalize in the next solver rust#136074)
  • tests/ui/implied-bounds/normalization.rs
  • tests/ui/implied-bounds/gluon_salsa.rs (region identity)
• implied bounds nyaaaaaaaaaa (properly supporting this needs canonicalization to map regions to universal instead of existential vars, matches old solver behavior now implied bounds: lifetime equality lost after normalization rust#106569)
  • tests/ui/implied-bounds/normalization-preserve-equality.rs

Handling both normalization while computing implied bounds and avoiding implied bounds from normalizing opaques is non-trivial and may require some more significant changes to the current design