Use constrain_build macro to simplify parallel make handling
With constrain_build it is possible to just say how much RAM per process
/ CPU core we need to build and it automatically calculates out the
right -jXX number to use.
I've used 2 GB per CPU as the initial setting, but if we run into issues
with gcc running out of memory on e.g. s390x, the number can be easily
increased (which then reduces parallelism on low memory builders).