Fix logic error introduced in #30334

[ahejlsberg]

This PR fixes a nasty little logic error I inadvertently introduced in #30334. This logic error is the cause of the DT failures reported in #30372.

I see that we've merged #30375 which doesn't actually fix the core issue. I don't like the fact that a bunch of circularity errors disappear with #30375 and I'd like for us to revert that PR.

[weswigham]

Err, I don't see it. What's logically different between the before and after here?

Anyways, you wanna revert the checker changes from #30375 but keep the test so we can see this fix the thing?

[ahejlsberg]

@weswigham I didn't see it either for quite a while. What's different is that when the !contains(...) check fails for contraCandidates (i.e. when the candidate is already in set of contravariant candidates), we proceed to the else clause which will add it to the covariant candidate set instead of doing nothing! And once we do that, who knows what happens. But the crazy thing is that it only goes wrong when we make two or more identical contravariant inferences.

[ahejlsberg]

@weswigham Yes, let's keep the test from #30375 but otherwise revert the PR.

[weswigham]

Ahhhhhh, I get it. Looking at the diff alone that's not obvious, lol.