Support single precision floats in grisu formatting

[orivej]

Currently fmt::format("{}", 0.1f) with grisu produces 0.10000000149011612 (as would have been expected from double(0.1f)) rather than 0.1.

Single precision formatting differs from double precision only in the calculation of the boundaries: https://github.com/google/double-conversion/blob/v3.1.5/double-conversion/fast-dtoa.cc#L525-L536

[ksharenkov]

As I see, floating format with "{}" and GRISU is not more compatible with std::defaultfloat in fmtlib 6.0.0

std::cout << fmt::format("{}\n", 1.0 / 3.0);
std::cout << 1.0 / 3.0 << "\n";

outputs

0.3333333333333333
0.333333

(fmtlib 6.0.0, GRISU, VisualStudio 2019 16.3)
It looks like compatibility problem.

[orivej]

Please use a separate issue to discuss that. #1033 is related, although it does not talk about the default cout format. {} is not supposed to output the same as cout <<, it is supposed to print the number with enough precision such that reading that number back produces the same number. 1/3 is most accurately approximated by a double 0.3333333333333333, it needs so many digits to distinguish it from the previous double 0.33333333333333326 and the next double 0.33333333333333337; 0.333333 represents a different number.

[vitaut]

Single precision formatting differs from double precision only in the calculation of the boundaries

Right. This is also needed for compatibility with std::format. @orivej, could you by any chance submit a PR?