Implement work-stealing queue approach to speed up parallel reduction

[FranchuFranchu]

A port of #42 to the new codebase.

[HigherOrderBot]

Perf run for a43dcfa:

file              mode        main          a43dcfa57c9d
========================================================
sum_rec           intr-singl      14.018 s      15.381 s
                  intr-multi       7.324 s       7.109 s
--------------------------------------------------------
bitonic_sort_lam  intr-singl      11.051 s      11.747 s
                  intr-multi       5.562 s       6.019 s
--------------------------------------------------------
merge_sort        intr-singl       7.387 s       7.567 s
                  intr-multi       3.660 s       4.142 s
--------------------------------------------------------
sum_tail          intr-singl       1.136 s       1.186 s
                  intr-multi       1.128 s       1.180 s
--------------------------------------------------------
c2                intr-singl       0.000 s       0.000 s
                  intr-multi       0.001 s       0.001 s
--------------------------------------------------------
sum_tree          intr-singl       9.113 s       9.837 s
                  intr-multi       4.314 s       4.789 s
--------------------------------------------------------
boom              intr-singl       2.232 s              
                  intr-multi       2.463 s       1.738 s
--------------------------------------------------------
radix_sort_lam    intr-singl       9.355 s      10.192 s
                  intr-multi       4.904 s       5.460 s