Make add/sub store all 64-bits of the result

The previous version only stored the lower 32-bits.
This meant that any results that would have
affected the upper 32-bits were ignored. This was
only a problem when jumps are fixed high in the
address space. But, definitely a big problem
nonetheless.