Rename InferCtxtLike to SolverDelegate, use a newtype to define an impl in rustc_next_trait_solver

[compiler-errors]

This is so we can add things like vtable walking methods to the SolverDelegate trait without needing to worry about how things are defined in rustc_trait_solver and not rustc_infer.

1. Move InferCtxtLike to rustc_next_trait_solver and rename it to SolverDelegate.
2. Add a newtype struct RustcSolverDelegate<'tcx>(InferCtxt<'tcx>) and use repr(transparent) to unsafely interconvert between them.

This is based on #126496 -- I split those relatively mechanical commits out in case we want to bikeshed on design here, since I'm not totally certain this is the best tradeoff of approach.

r? lcnr

[rustbot]

Some changes occurred to the core trait solver

cc @rust-lang/initiative-trait-system-refactor

[compiler-errors]

This might be overkill. I could perhaps alternatively uplift the vtable walking code into rustc_next_trait_solver, and keep InferCtxtLike as is. AFAICT there's no other places in the new solver that rely on code that's implemented in rustc_trait_selection not rustc_infer, but I'm not certain if it'll stay that way.

edit: Actually, specialization_graph machinery is defined in rustc_trait_solver. However, we could also emulate that by just exposing the Ancestors iterator.

[compiler-errors]

Actually, yeah? I think this is overkill lol. If the only reason we need to do this is because of walk_vtable, then I'd rather find a better solution for that.

[compiler-errors]

the better solution: #126505