Better handling of partial sums

[richfitz]

Consider this from sircovid

n_E_vaccinated[, ] <- sum(n_E_next_vacc_class[i, , , j])
dim(n_E_vaccinated) <- c(n_groups, n_vacc_classes)

this compiles to

    for (int i = 1; i <= shared->dim_n_E_vaccinated_1; ++i) {
      for (int j = 1; j <= shared->dim_n_E_vaccinated_2; ++j) {
        internal.n_E_vaccinated[i - 1 + shared->dim_n_E_vaccinated_1 * (j - 1)] = odin_sum4<real_t>(internal.n_E_next_vacc_class.data(), i - 1, i, 0, shared->dim_n_E_next_vacc_class_2, 0, shared->dim_n_E_next_vacc_class_3, j - 1, j, shared->dim_n_E_next_vacc_class_1, shared->dim_n_E_next_vacc_class_12, shared->dim_n_E_next_vacc_class_123);
      }
    }

We should be able to do this more efficiently, depending on how the indices here are being summed over, which might result in less churn through the loops here and a more efficient sum.