Often I study some JavaScript interview questions, suddenly I saw a question about usage of reduce function for sorting an Array, I r
reduce constraints yourself to online sorting algorithms, where each element in the array is seen once and you do not now in advance the length of your array (note that using closures you could give more information to the reducing function but this kinds of defeats the purpose of the question).
insertion sort is an obvious and easy example; I won't detail the implementation as the other answers are already very good in this regard. However you can mention a few optimizations that might probably seen as very positive to the interviewer:
Use binary search to find the insertion point can reduce the complexity of an insertion step from O(n) to O(log n). This is called binary insertion sort: the overall number of comparisons will go from O(n^2) to O(n log n). This won't be faster because the real cost is due to "splicing" the output array but if you had expensive comparisons (say, sorting long strings for instance) it could make a difference.
If you sort integer, you can use radix sort and implement a linear online sorting algorithm with reduce. This is quite trickier to implement, but very suited to a reduction.