Fix: -Wfloat-conversion error in print_number on MinGW

[coveyjorjet]

Closes #963

Problem

On MinGW-w64, isnan/isinf in math.h are macros that dispatch to float-typed helpers (__isnanf, __fpclassifyf, ...) via __builtin_choose_expr. Passing a double then triggers:

error: conversion from 'double' to 'float' may change value [-Werror=float-conversion]

when building with -Wconversion/-Wfloat-conversion + -Werror (as cJSON's own CMake custom flags do).

Fix

Replace the isnan(d) || isinf(d) call in print_number with a pure C89 IEEE-754 check that only operates on double:

if ((d != d) || (d > DBL_MAX) || (d < -DBL_MAX))

• d != d is true exactly for NaN
• d > DBL_MAX / d < -DBL_MAX are true exactly for +/-Inf (DBL_MAX is the largest finite double)

This avoids the platform isnan/isinf macros entirely, so no double-to-float conversion warning is generated. It is also fully ANSI C (C89), so it works on old compilers such as GCC 2.95 that do not provide __builtin_isnan/__builtin_isinf, and on non-GCC compilers that define __GNUC__.

The now-unused ANSI C fallback isnan/isinf macros were also removed (they would otherwise trigger -Wunused-macros on platforms where they get defined).

Verification

• Compiles clean with cJSON's strict flags: -std=c89 -pedantic -Wall -Wextra -Werror -Wconversion -Wfloat-conversion -Wunused-macros ...
• make test passes
• All 20 standalone test binaries pass
• NaN, +Inf, -Inf print as null; normal numbers unchanged

[wissem01chiha]

Thanks very much @coveyjorjet , for the PR , do you have any idea why it failed with GNU 15.2.0 and worked with GNU 15.3.0 ? or maybe i have something in my local setup that overlap like includes from other compilers (i MSVC and clang installed) ? thanks anyway