Introduce SolverRelating type relation to the new solver code

[compiler-errors]

This "uplifts" (i.e. duplicates) the TypeRelating delegate into the new solver as a new relation called SolverRelating, so we can use that rather than exposing InferCtxtLike::relate for implementations to have to implement themselves.

The reason I duplicated rather than fully uplifted TypeRelating is because it's so complicated.

• CombineFields is a bunch of unnecessary machinery that only exists to track a Trace and to make it easy to swap between Lub/Glb.
• There are a bunch of next_trait_solver branches in in the super_combine_{const,ty} functions, and in the TypeRelating code itself that can be folded away.

This then allows us to remove StructurallyRelateAliases from PredicateEmittingRelation and move it directly onto the generalizer instantiate_*_var functions.

r? @lcnr -- I'm curious what you think of this approach in general.

[rustbot]

changes to the core type system

cc @compiler-errors, @lcnr

[lcnr]

generally in favor of splitting TypeRelating between old and new solver, have a bunch of vibes about how to do that though, probably useful to chat about them in sync though

[lcnr]

we discussed that we'll have to add InferCtxtLike::next_trait_solver anyways to uplift the generalizer, at which point I strongly preferred to uplift super_combine instead of duplicating the type relations.

[lcnr]

@rustbot author

[bors]

☔ The latest upstream changes (presumably #129365) made this pull request unmergeable. Please resolve the merge conflicts.

[compiler-errors]

I'll need to totally rework this, so closing for now and putting it back onto my TODO list for a fresh PR, which will be soon hopefully.