+Add optional unscale argument to reproducing_sum

  Added the new optional unscale argument to reproducing_sum() and
reproducing_sum_EFP().  When this is used, the resulting real sum is restored to
the same scaling as the input arrays, but when there is an EFP_type returned it
is the same as it would have been had the input array been rescaled before it
was passed in.  All answers are bitwise identical, but thre is a new optional
argument to a two publicly visible interfaces.