Lower the threshold to use sequential sort

The threshold lowered from 64 elements to sort, to 16 elements.
This recovers monotonicity of run-times curve.

When number of array elements along the axis we sort is less
than the threshold, a quadratic complexity insertion sort
is used. Too high a value of the threshold (64) made for the
artifact, where sorting 63 elements took several times longer
than sorting 65 elements.

Setting it to 16 recovers monotonicity (sorting 16 elements
is not slower than sorting 17 elements).

Author: oleksandr-pavlyk

dpctl/tensor/libtensor/include/kernels/sorting/sort.hpp

@@ -734,7 +734,7 @@ sycl::event stable_sort_axis1_contig_impl(
 
     auto comp = Comp{};
 
-    constexpr size_t sequential_sorting_threshold = 64;
+    constexpr size_t sequential_sorting_threshold = 16;
 
     if (sort_nelems < sequential_sorting_threshold) {
         // equal work-item sorts entire row