<cmath>: floating-point classification functions are ambiguous for integer types

[frederick-vs-ja]

Currently calling a floating-point classification/comparison function in <cmath> with an integer value as argument such as std::isnan(42) is ambiguous. And the standard says that sufficient overloads are provided for integer types ([cmath.syn]/2, at least since N3337), so the ambiguity should be resolved.

Overloads of other <cmath> functions for integer types are handled in <cmath>. So overloads of these functions should also be handled in this header.

[StephanTLavavej]

Duplicate of #519.