Replace OMP reductions by thread-private variables + critical

[marvinfriede]

Replace reductions on arrays, which can lead to stack overflow errors for large(r) arrays. Reductions on arrays usually require increasing both OMP_STACKSIZE for the worker threads, but also setting ulimit -s unlimited to allow the stack size of the main thread to grow beyond the limit set upon starting the process. This is because reductions put the arrays on the stack, not on the heap.
We avoid this by creating private copies on the heap for each thread and add up all copies after the loop in a critical section, effectively emulating the reduction.