hex FP constants

[David G. Hough on validgh]

In general, hex FP constants are intended to promote portability among IEEE
implementations of portable values, and as proposed satisfy this criterion
except for infinity, since NaN semantics aren't portable.    hex FP constants
aren't portable in the intended way,
among all existing floating-point implementations, but I consider that 
acceptable: the most important benefit of FPCE is providing the language
bindings that were missing from IEEE 754, and anything that works as well
for non-754 implementations is just added gravy, without implying that 
things that don't work for non-754 implementations shouldn't be specified
at all.    Bit patterns will still be used from time to time even on IEEE
implementations, but the hex FP constants as proposed satisfy 90% of the
problem in a clean way.