[sprintf] Wrong formatting of nan/inf with leading 0

[stefano-zanotti-88]

In a footnote, the standard says:

When applied to infinite and NaN values, the -, +, and space flag characters have their usual meaning; the # and 0 flag characters have no effect.

STB vs standard-conforming implementation (notice also the wrong upper/lower case, already reported):

printf("%f", STB_POS_QNAN);   // "NaN"    vs "nan"
printf("%f", STB_NEG_QNAN);   // "-NaN"   vs "-nan"
printf("%f", STB_POS_SNAN);   // "NaN"    vs "nan"
printf("%f", STB_NEG_SNAN);   // "-NaN"   vs "-nan"
printf("%f", 0/0.0);          // "-NaN"   vs "-nan" or "nan" (the sign of nans is not guaranteed)
printf("%f", +1/0.0);         // "Inf"    vs "inf"
printf("%f", -1/0.0);         // "-Inf"   vs "-inf"
printf("%06f", STB_POS_QNAN); // "000NaN" vs "   nan"
printf("%06f", STB_NEG_QNAN); // "-00NaN" vs "  -nan"
printf("%06f", STB_POS_SNAN); // "000NaN" vs "   nan"
printf("%06f", STB_NEG_SNAN); // "-00NaN" vs "  -nan"
printf("%06f", 0/0.0);        // "-00NaN" vs "  -nan"
printf("%06f", +1/0.0);       // "000Inf" vs "   inf"
printf("%06f", -1/0.0);       // "-00Inf" vs "  -inf"

Reproducible nans for testing can be generated this way:

double make_nan(int sign, int signaling)
{
    unsigned long long v = ((sign ? 1ull : 0ull) << 63)
        | (0x7FFull << 52)
        | (signaling ? 1ull : (1ull << 51))
        ;
    double d;
    memcpy(&d, &v, 8);
    return d;
}
#define STB_POS_QNAN    make_nan(0, 0)
#define STB_NEG_QNAN    make_nan(1, 0)
#define STB_POS_SNAN    make_nan(0, 1)
#define STB_NEG_SNAN    make_nan(1, 1)

Fix: in both the doexpfromg: and dofloatfromg: code paths, add this:

if (dp == STBSP__SPECIAL) {
   fl &= ~STBSP__LEADINGZERO;
   s = (char *)sn;
   cs = 0;
   pr = 0;
   goto scopy;
}