[Bug tree-optimization/118660] [14/15 Regression] VRP gets in the way sometimes

[pinskia at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118660

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
This one is slightly harder to handle:
```
unsigned short f(unsigned a)
{
  if ((a & 3) != 3) return a & 1;
  return a & 1;
}
```

We get:
```
  <bb 2> [local count: 1073741824]:
  _1 = a_4(D) & 3;
  if (_1 != 3)
    goto <bb 3>; [50.00%]
  else
    goto <bb 4>; [50.00%]

  <bb 3> [local count: 536870912]:
  _2 = (short unsigned int) a_4(D);
  _5 = _2 & 1;

  <bb 4> [local count: 1073741824]:
  # _3 = PHI <_5(3), 1(2)>

```

I suspect for this one, we need to add an extra code to phi-opt to allow for an
optional conversion too.

NOTE I would say this should be deffered until GCC 16 since now both GCC and
LLVM have the same issue and I doubt it was noticed in real code.