"Fallback" to i32/f32 in error messages involving numeric inference

[Aatch]

Currently error messages where the type of a variable is a integer variable or a float variable just put _ in for the type like a regular inference variable. For straight type mismatch errors, the elaborated version prints out something like an integral variable, which is fine, but other error messages don't. Like trait core::ops::Div<_> is not implemented for f64.

Given that we have fallbacks to i32 and f32 for integer and float literals respectively, it might be worth error messages acting as if that fallback has happened. trait core::ops::Div<i32> is not implemented for f64 is much more helpful, since it's fairly clear that the issue is that you're dividing a float by an integer.

[pnkfelix]

dupe of #24770 ?

[alexcrichton]

I'd agree this is a dupe.