Relaxing checks for MULT16_32_QX()

MULT16_32_QX() is now implemented using a signed-unsigned multiply,
so the second argument can now have one extra bit compared to the
old signed-signed implementation.

Reviewed by Mark Harris