test(h5dump): reduce printf precision for t128bit_float

[hyoklee]

H5DUMP-t128bit_float has a long double width portability issue (E.g., Arm64 OpenBSD).
The 80-bit (x86) and 128-bit (arm64) renderings agree to 20 sig-figs then diverge.

Possible solution: dump at %.18Lg (portable across widths, still proves long
double > double) and update the .ddl.