rustc should not try to analyze the crate after "multiple candidates" messages

[RalfJung]

#118129 is an example of an issue that somehow leads to "multiple candidates" errors for each rustc crate. However, it took hours to get to that conclusion, because rustc shows >1800 errors saying all sorts of things could not be found. The actual signal is drowning in that noise. This is also fairly common when something goes wrong with the sysroot and it has multiple core/std candidates and imports none of them: all one can see are hundreds of errors saying that some lang item is missing.

It would be better if rustc stopped after crate loading when there was such a "multiple candidates" error, or at least didn't show all the follow-up messages.

[bjorn3]

I think we should abort after failing to load libcore or libstd in general independent of whatever error caused it.