[Patches] fix hashing take#2 (was: fix float_hash and complex_hash for 64-bit *nix)

[Tim Peters]

[Trent Mick]
> ...
> Tim, a question: Before, the float_hash algorithm that we were
> discussing did the frexp-and-32-bits-at-a-time reduction magic to
> the fractpart of the float and then you identified that the same
> needed to be done to the intpart.
> Instead of doing this twice, I just applied the reduction to
> whole double (fractpart+intpart). I don't see any error in doing that.
> Do you?  As well, is there some hashing benefit that is lost by
> my doing the reduction only once?

No, no problem, looks good!  The simplification makes up for the extra cost
of the function call, too <wink>.