Fix implicit conversion of int64 to int

[NicolasHug]

Closes #679

[NicolasHug]

Not a fan creating a new cpp file for this, but I'm not sure where to have it otherwise. It's needed for both the pybind_ops and the custom_ops. Happy to consider alternatives.

Also, instead of listing ValidationUtils.cpp for both the custom_ops and pybind_ops libs, we could just list it once as part of the core lib.

[Dan-Flores]

It makes sense to create a new cpp file for this, we may add more validations functions in the future.

Since it is used in both libraries I think it makes sense to add ValidationUtils.cpp to the core lib, as the other shared libraries are also there (SingleStreamDecoder.cpp, Encoder.cpp).

[scotts]

A new cpp for this is the best practice, as things are currently implemented. However, we might end up removing the cpp in the future because if we find ourselves implementing variants of these validation functions, we should probably instead define something like:

template <SrcT, DestT>
DestT validateIntegerConversion(SrcT value, const std::string& parameterName) {
  TORCH_CHECK(
      value >= std::numeric_limits<DestT>::min() &&
          value <= std::numeric_limits<DestT>::max(),
      parameterName,
      "=",
      value,
      " is out of range for int type.");

  return static_cast<DestT>(value);
}

And the definition of template functions must live in the header files. I don't think we should make this change yet because:

1. My code above may not generalize correctly! Testing it's general correctness is not worth it right now.
2. We may never need anything beyond this specific validation of types. (YAGNI.)

Yeah, I didn't do this in #830, but it was one function, and one could argue I was just being lazy. :)