Restore add(dp_op) to operator chain, remove stale contributeHk call

The previous commits removed add(dp_op) as a workaround for np=2 crash,
breaking np=1 constraint mode. Restored.

The MPI fixes that remain valid:
- dp_target broadcast (bcast_cell.cpp)
- Gamma Bcast (deltap_wannier.cpp)
- Lambda Bcast (esolver_ks_lcao.cpp)
- Print rank-0 guards (esolver_ks_lcao.cpp)
- Null guard in deltap_lcao.cpp

Author: dyzheng

source/source_esolver/esolver_ks_lcao.cpp

@@ -1411,19 +1411,6 @@ void ESolver_KS_LCAO<TK, TR>::deltap_update_lambda(UnitCell& ucell, const int it
         std::unordered_map<int, std::vector<std::complex<double>>> hk_corr;
         dp->compute_hk_correction(ucell, psi, lambda, hk_corr);
         dp_op->set_hk_correction(hk_corr);
-
-        // Apply HK correction on rank 0 only (operator not in chain for MPI)
-#ifdef __MPI
-        if (this->pv.comm() != MPI_COMM_NULL) {
-            int rank = 0;
-            MPI_Comm_rank(this->pv.comm(), &rank);
-            if (rank == 0)
-#endif
-                for (int ik = 0; ik < this->kv.get_nks(); ++ik)
-                    dp_op->contributeHk(ik);
-#ifdef __MPI
-        }
-#endif
     }
 }
 

source/source_lcao/hamilt_lcao.cpp

@@ -417,12 +417,10 @@ HamiltLCAO<TK, TR>::HamiltLCAO(const UnitCell& ucell,
                                                               &grid_d,
                                                               two_center_bundle.overlap_orb_onsite.get(),
                                                               orb.cutoffs(),
-                                                              PARAM.inp.deltap_rm);
+                                                               PARAM.inp.deltap_rm);
+            this->getOperator()->add(dp_op);
             this->dp_operator = dynamic_cast<DeltaPOperator<TK, TR>*>(dp_op);
             this->dp_operator->set_gdir(PARAM.inp.deltap_gdir);
-            // Note: NOT added to operator chain (getOperator()->add) due to
-            // MPI distributed-k-point crash. HK correction applied directly
-            // from esolver iter_finish.
         }
     }