bpo-45440: Require math.h isinf() to build (GH-28894)

Building Python now requires a C99 <math.h> header file providing
isinf(), isnan() and isfinite() functions.

Remove the Py_FORCE_DOUBLE() macro. It was used by the
Py_IS_INFINITY() macro.

Changes:

* Remove Py_IS_NAN(), Py_IS_INFINITY() and Py_IS_FINITE()
  in PC/pyconfig.h.
* Remove the _Py_force_double() function.
* configure no longer checks if math.h defines isinf(), isnan() and
  isfinite().

Python/pymath.c

@@ -1,18 +1,5 @@
 #include "Python.h"
 
-#ifdef X87_DOUBLE_ROUNDING
-/* On x86 platforms using an x87 FPU, this function is called from the
-   Py_FORCE_DOUBLE macro (defined in pymath.h) to force a floating-point
-   number out of an 80-bit x87 FPU register and into a 64-bit memory location,
-   thus rounding from extended precision to double precision. */
-double _Py_force_double(double x)
-{
-    volatile double y;
-    y = x;
-    return y;
-}
-#endif
-
 
 #ifdef HAVE_GCC_ASM_FOR_X87
 // Inline assembly for getting and setting the 387 FPU control word on