Fix a lapse in logic for computing upper bound of iteration

This fixes gh-1901

Author: oleksandr-pavlyk

dpctl/tensor/libtensor/include/kernels/accumulators.hpp

@@ -281,7 +281,7 @@ inclusive_scan_base_step(sycl::queue &exec_q,
             }
 
             const nwiT m_max =
-                std::min<nwiT>(n_wi, std::max(i, acc_nelems) - i);
+                static_cast<nwiT>(std::min(i + n_wi, acc_nelems) - i);
             for (nwiT m_wi = 0; m_wi < m_max; ++m_wi) {
                 output[out_iter_offset + out_indexer(i + m_wi)] =
                     local_iscan[m_wi];