[InstCombine] Missed optimization for add + and + icmp pattern

https://alive2.llvm.org/ce/z/7A_uJr

```llvm
define i1 @src(i8 range(i8 0, 4) %arg) {
  %add = add nsw i8 %arg, -1
  %and = and i8 %add, -3
  %cmp = icmp eq i8 %and, 0
  ret i1 %cmp
}

define i1 @tgt(i8 range(i8 0, 4) %arg) {
  %and = and i8 %arg, 1
  %cmp = icmp ne i8 %and, 0
  ret i1 %cmp
}
```

Not sure how to generalize this best.