Fix enpt2

[peter-reinholdt]

There was a problem where the occupations in t_up did not match det_up in the 1-1 excitation elements.

Originally, the 1-0 branch had the following bit of code

            if (std::abs(val) > eps) {
                rank = wfn.rank_det(det_up);
                if (wfn.index_det_from_rank(rank) == -1) {
                    val *= sign_up;
                    fill_occs(wfn.nword, det_up, t_up);
                    compute_enpt2_thread_gather(wfn, ham.one_mo, ham.two_mo, terms[rank], val, n2,
                                                n3, t_up);
                }
            }

which would assign t_up for the 1-0 branch as needed, i.e., when |Hki*ci|>eps and when the singly-excited determinant was not contained in the variational wave function.

However, t_up needs to be correctly assigned also for the later 1-1 branch, which would not always happen.

Additionally, the threads are not guaranteed to join in order, so the compute_enpt2_thread_condense now happens only after all threads have joined.

[msricher]

Thanks, will check it out and merge

[msricher]

yes, seems you are right