Fix load/save to handle up to 2^32-1 elements. Use std::streamsize in calculations of number of bytes, rather than octave_idx_type, so that expressions like 8 * nel do not overflow. * ls-mat4.cc (save_mat_binary_data): Use std::streamsize in calculations of number of bytes. * ls-mat5.cc (read_mat5_integer_data, MAT5_DO_WRITE): Use std::streamsize in calculations of number of bytes. * data-conv.cc (LS_DO_READ, LS_DO_READ, read_doubles, read_floats, write_floats): Use std::streamsize in calculations of number of bytes.