(FLO_union_type): Remove bitfields to set sign, exponent, and
mantissa, and add value_raw field, which is an integer of the
appropriate type. If _DEBUG_BITFLOAT is defined, provide little and
big endian bitfields.
(pack_d, unpack_d): Switch to use value_raw and explicit shifts and
masks so that we don't have to worry about whether the target is big
or little endian. If single precision floating point, rename to
pack_f and unpack_f, so there is no confusion in the debugger.
From-SVN: r10312
446c894728