[PSA] Swapping out `_or_null` with `_if_present`

[nikic]

@bzcheeseman

All types that go into if_present go through a struct called ValueIsPresent and we actually handle raw pointers, smart pointers, and types like mlir::Type through the same overload (they’re all ‘nullable’), so that’s definitely part of my reticence to split it up into 2 different things. That said, IMO it’s not great to have 2 things that do conceptually the same thing (check for existence, then cast) in slightly different ways. That feels like a recipe for code drift/maintainability problems.

I think the suggestion here is to just limit the functions using the “null” terminology to overloads accepting pointers. So if we currently have

template <class X, class Y> auto cast_or_null(const Y &Val) {
  return cast_if_present<X>(Val);
}

template <class X, class Y> auto cast_or_null(Y &Val) {
  return cast_if_present<X>(Val);
}

template <class X, class Y> auto cast_or_null(Y *Val) {
  return cast_if_present<X>(Val);
}

template <class X, class Y> auto cast_or_null(std::unique_ptr<Y> &&Val) {
  return cast_if_present<X>(std::move(Val));
}

then we can reduce this to just

template <class X, class Y> auto cast_or_null(Y *Val) {
  return cast_if_present<X>(Val);
}

This limits the use of cast_or_null (and friends) to just the case where the naming makes sense.

Also worth noting that next to the two functions already discussed we also have isa_and_nonnull / isa_and_present. The former is self-explanatory, while the latter is not.