Rollup merge of #144711 - compiler-errors:op-span, r=petrochenkov

Consider operator's span when computing binop expr span

When computing the span of a binop consisting of `lhs` and `rhs`, we previously just took the spans of `lhs.span.to(rhs.span)`. In the case that both `lhs` and `rhs` are both arguments to a macro, this can produce a wildly incorrect span.

To fix this, first compute the span between `lhs` and the binary operator, which will cause `lhs` to possibly be adjusted to a relevant macro metavar, and then compute that span extended to `rhs`, which will cause it to also be adjusted to a relevant macro metavar.

This coincidentally fixes a FIXME in `tests/ui/lint/wide_pointer_comparisons.rs` and suppresses a nonsense suggestion.

Author: jdonszelmann