Ensure sort_crs_matrix in Kokkos is only performed on matrices with local column indices

[stevendargaville]

Summary

In remove_small_from_sparse_kokkos, KokkosSparse::sort_crs_matrix was called on the nonlocal block while j_nonlocal_d still held global column indices, but the KokkosCsrMatrix was constructed with numCols = cols_ao (the local count). Global indices >= cols_ao can cause out-of-bounds access inside the sort.

• Split rewrite_j_global_to_local into device-side helpers (rewrite_j_global_to_local_device + remap_j_to_local_device, reusing the existing binary_search_sorted device function) with the host garray copy kept in the original wrapper, whose signature is unchanged for its other callers.
• In remove_small_from_sparse_kokkos, remap j_nonlocal_d to local indices before the lumping sort, and construct csrmat_nonlocal with the correct col_ao_output as numCols. The output block now just copies the device garray to the host.

MatAXPY_kokkos needs no change: it already performs the spadd in a merged-local column space.

Testing

• Warning-clean build
• make check and make tests_short pass: plain, with -mat_type aijkokkos -vec_type kokkos -dm_mat_type aijkokkos -dm_vec_type kokkos, and with PFLARE_KOKKOS_DEBUG=1

🤖 Generated with Claude Code