Use float32/64 instead of float/double

Use float32/64 instead of float/double

The patch below uses the float32 and float64 types instead of the float
and double types in the PPC code. This doesn't change anything when
using softfloat-native as the types are the same, but that helps
compiling the PPC target with softfloat.

It also defines a new union CPU_FloatU in addition to CPU_DoubleU, and
use them instead of identical unions that are defined in numerous
places.

git-svn-id: svn://svn.savannah.nongnu.org/qemu/trunk@4047 c046a42c-6fe2-441c-8c8c-71466251a162