Avoid synthetic auto-trait ICE with next solver in rustdoc

[TaKO8Ki]

Fixes #156487

AutoTraitFinder still uses old-solver-only selection APIs. When rustdoc runs with -Znext-solver=globally, synthetic auto-trait impl generation reaches that path with the new solver enabled and ICEs in trait selection.

[rustbot]

r? @camelid

rustbot has assigned @camelid.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

Why was this reviewer chosen?

The reviewer was selected based on:

• Owners of files modified in this PR: rustdoc
• rustdoc expanded to 9 candidates
• Random selection from GuillaumeGomez, camelid, fmease, lolbinarycat, notriddle

[camelid]

Could you explain a bit more about the purpose of this PR? It seems like it just turns a rustdoc ICE into silently producing incorrect results with -Znext-solver, which doesn't seem good to me. Is it to test the other parts of rustdoc with -Znext-solver? Apologies if I'm missing some context.

View changes since this review

[TaKO8Ki]

@camelid

Sorry, I opened this PR while I was still debugging and before I had completed the changes. I think it is better for rustdoc to return a normal error with fatal as an example for now, e.g.

rustdoc does not generate auto-trait impls for display with -Znext-solver=globally

rather than causing an ICE.

What do you think?

[camelid]

I don't think we want to just get rid of this ICE or turn it into an error since it's just due to missing code in rustdoc. I.e., we need to just add code that uses the new solver. Please let me know if I'm misunderstanding. Closing in favor of #149019.