mp_snprint() should preserve float values precision

[parihaaraka]

huge precision loss of fiber.time() values :(

[Totktonada]

We can replace %g / %lg with something like %.17g / %.17lg.

However it'll not guarantee that the precision will not loss. There are two points here:

• %.17lg (unlike %.17lf) may choose the scientific notation for a large value. So, 111<...>111 will become 1.111<...>111e+X and the precision (17) will be applied to the 'fractional part' of the normalized value, not the original one.
• There are doubles, which becomes near to a halfway between two exact doubles, after converting to the decimal representation. Marshalling over decimal may give us back not exactly same value. See https://www.exploringbinary.com/17-digits-gets-you-there-once-youve-found-your-way/

Converting to hex would be good option for cases, when exact value should be stored.

[parihaaraka]

Most users understand that binary floating point value may not preserve its original precision via decimal representation. "huge" means HUGE precision loss - out of acceptable dec<->bin transformations impact. Actually mp_snprint() must not be used for floating point numbers at all.