fix(dftu): guard force/stress against non column-major ks_solver

The folded dSR/mat buffers are consumed by ScaLAPACK GEMM via pv.desc
(local column-major storage) and read back with explicit ic * pv.nrow +
ir indices, so their layout must not follow a row-major ks_solver
branch. All LCAO ks_solvers accepted by INPUT validation satisfy
IS_COLUMN_MAJOR_KS_SOLVER today, so this entry guard is defensive: it
fails loudly instead of silently producing wrong DFT+U forces/stresses
if solver whitelists ever drift.

Notes on governance warnings:
- Tests not added: the guarded behavior is a fail-fast precondition on
  a path whose numerical results are unchanged for every reachable
  configuration; dft_plus_u==2 force/stress is additionally disabled in
  force_stress_lcao.cpp. A focused unit test can be added together with
  the planned follow-up that fixes folding layout at construction time.
- No docs update required: no INPUT parameter semantics change.

Author: abacus_fixer

source/source_lcao/module_dftu/dftu_force.cpp

@@ -74,6 +74,19 @@ void force_stress(Plus_U& dftu,
         }
     }
 
+    // Layout invariant: the folded dSR/mat buffers are consumed by ScaLAPACK
+    // GEMM through pv.desc (local column-major storage) and read back with
+    // explicit ic * pv.nrow + ir indices. All ks_solvers accepted by INPUT
+    // validation are column-major today; abort loudly instead of silently
+    // producing wrong forces/stresses if that assumption ever changes.
+    if ((dftu.is_cal_force() || dftu.is_cal_stress())
+        && !ModuleBase::GlobalFunc::IS_COLUMN_MAJOR_KS_SOLVER(dftu.get_ks_solver()))
+    {
+        ModuleBase::WARNING_QUIT("DFTU_LCAO::force_stress",
+            "non column-major ks_solver is not supported for DFT+U force/stress; "
+            "the folded matrix layout assumption would be violated");
+    }
+
     const int nlocal = dftu.get_nlocal();
 
     if (dftu.is_cal_force())