fix nvhpc compilation

[simonpintarelli]

nvhpc crashed with a segfault on hamiltonian_k.cpp when using sirius+cuda. Most likely cause is the number of template instantiations/TU complexity.

• remove the declaration for omp reduction for std::complex when nvhpc is used (this gave a compilation error)
• rewrite (reorder) omp for loops in hamiltonian_k.cpp to avoid the segfault
• rewrite range-based for loops (nvhpc seems to require the canonical form)

The error for range-based for loops with nvhpc are the following:

NVC++-S-0155-Unsupported initialization in OMP for loop  (rangefor_omp_test.cpp: 24)
NVC++-S-0155-#pragma pfor does not match canonical form  (rangefor_omp_test.cpp: 24)
NVC++/arm Linux 26.5-0: compilation completed with severe errors

[simonpintarelli]

cscs-ci run default

[simonpintarelli]

CI on eiger is failing, the base uenv on eiger requires a rebuild

[simonpintarelli]

cscs-ci run default

[simonpintarelli]

cscs-ci run default

[simonpintarelli]

cscs-ci run default

[toxa81]

One comment about missing OMP statement in the hamiltonian_k.cpp and more concerning issue: my PR #1168 might cause a lot of conflicts. I will probably rebase it on top of this one.

[toxa81]

What about OMP statement over atoms? In case of CPU this will be executed sequentially atom by atom and omp parallelization will only come via threaded blas

[simonpintarelli]

True, I forgot about the CPU only case. I'll revert this change.

[simonpintarelli]

I've reverted the loops over the atoms to using openmp, compilation with nvhpc is still working.

[simonpintarelli]

cscs-ci run default