Change to pass sorting direction as call argument, not template parameter

The intent is to reduce the build time, build memory footprint, and binary size of the sorting_impl module. With this change it stands at 46MB, before it was 72MB.

Author: oleksandr-pavlyk

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

@@ -807,8 +807,7 @@ sycl::event stable_argsort_axis1_contig_impl(
  const IndexComp<IndexTy, argTy, ValueComp> index_comp{arg_tp, ValueComp{}};
 
  static constexpr size_t determine_automatically = 0;
- size_t sorted_block_size =
- (sort_nelems >= 512) ? 512 : determine_automatically;
+ size_t sorted_block_size = determine_automatically;
 
  const size_t total_nelems = iter_nelems * sort_nelems;