Intermediate overflows are in some situations allowed

[TomAFrench]

There's a number of cases where due to simplifications on operations such as x^x or x-x we allow intermediate overflows.

This came up recently in #7189 where the program below started to pass due to being able to simplify var1 to 0 and then removing all the overflows through dead instruction elimination.

#[test(should_fail_with = "attempt to add with overflow")]
fn main() {
    let var1: u8 = ((255 + 1) ^ (255 + 1)) - ((255 + 1) - (255 + 1));
    assert_eq(var1, 0);
}

We should pay attention to whether we want to perform dead instruction elimination in these cases or if we should preserve the addition/subtractions at least to maintain checked arithmetic.