You can already sort with couting-sort using numpy: as stated in an SO answer, it is simply:

np.repeat(np.arange(1+x.max()), np.bincount(x))

But there is no way (as far as I can tell) to do an argsort.

I think it would be fairly easy to implement and slot into the argsort/sort documentation. The benefit of counting sort is that it's O(n), although it only works for not-too-large non-negative integers...however those are actually quite common in the real world! Pandas uses it behind the scenes apparently.