Convert the MOVE macro on x86-64 into a function.
This is possible as the macro is used just to simulate an inline
function. Converting MOVE into a true function shrinks the core by
448 KiB and may even make the compiler run faster due to reduced
instruction cache pressure.
Some background: Only on x86-64 MOVE is used with float SCs sometimes.
It therefore needs to select different machine instructions depending on
the SC of its destination argument. This compiles to so much code that
inlining it can't be justified, especially given that MOVE is used in
several hundred VOPs.
While at it, correct the comment at the top of the file for 64-bitness.