Improve (format "%g" bignum) precision

Improve (format "%g" bignum) precision

* src/editfns.c (styled_format): When formatting bignums with
floating-point conversions like %g, use long double if that
would lose less information than double, which is what the
code was already doing for fixnums.  On Fedora 28 x86-64, for
example, (format "%.100g" (1- (ash 1 64))) now yields
"18446744073709551615" instead of the numerically incorrect
"18446744073709549568".  Also, fix a stray INTEGERP that
can just be FIXNUMP, since bignums are not possible there.