use radix sort when the comparator is just a key mapping

[pca006132]

Radix sort is faster, but we are not using it as much as possible due to some sort operations requiring custom comparators. We can generalize the radix sort implementation to allow a key mapping function.

The merge function is probably something we can look into as well, it seems that the std implementation is not doing much optimization with it.

[pca006132]

And I forgot why the radix sort is not really a parallel radix sort. I should fix thid...