Use MPI_Dist_graph_create to determine comm pattern

[aprokop]

MPI_Alltoall has issues with scaling. This patch replaces it by using MPI_Dist_graph_create and other helper functions to rely on MPI. This seems to scale better on Frontier:

This bumps our MPI requirement to 2.2.

Also replaced OpenMPI with MPICH in the HIP build.

@dalg24 pointed out that we did something similar in ORNL-CEES/DataTransferKit#516. So this can be seen as a partial revert of that.

[aprokop]

Tested it on Frontier. Need to test it more thoroughly:

• on Aurora to see if it helps with scaling
• using GPU-aware MPI

[Rombur]

Can we save the graph or do we need to rebuild it every time?

[aprokop]

Right now, we don't use it past Distributor construction. I do want to try to use it for neighborhood communications inside doPostsAndWaits in the future. I'll store it then if it works out.

[aprokop]

Replaced OpenMPI with MPICH in the HIP build.

[aprokop]

Switching HIP build from OpenMPI to MPICH results in very weird errors when building ArborX:

>>>               CMakeFiles/ArborX_Example_HostAccessTraits.exe.dir/example_host_access_traits.cpp.o:(Kokkos::Impl::SharedAllocationRecord<void, void>* Kokkos::Impl::ViewMapping<Kokkos::ViewTraits<int*, Kokkos::Device<Kokkos::OpenMP, Kokkos::HostSpace>>, void>::allocate_shared<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char>>, Kokkos::HostSpace, Kokkos::OpenMP>(Kokkos::Impl::ViewCtorProp<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char>>, Kokkos::HostSpace, Kokkos::OpenMP> const&, Kokkos::LayoutRight const&, bool))
>>> referenced by Kokkos_SharedAlloc.hpp:474 (/opt/kokkos/include/impl/Kokkos_SharedAlloc.hpp:474)
>>>               CMakeFiles/ArborX_Example_HostAccessTraits.exe.dir/example_host_access_traits.cpp.o:(Kokkos::Impl::SharedAllocationRecord<Kokkos::HostSpace, Kokkos::Impl::ViewValueFunctor<Kokkos::Device<Kokkos::OpenMP, Kokkos::HostSpace>, int>>::SharedAllocationRecord<Kokkos::OpenMP>(Kokkos::OpenMP const&, Kokkos::HostSpace const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char>> const&, unsigned long))
>>> referenced by Kokkos_SharedAlloc.hpp:482 (/opt/kokkos/include/impl/Kokkos_SharedAlloc.hpp:482)
>>>               CMakeFiles/ArborX_Example_HostAccessTraits.exe.dir/example_host_access_traits.cpp.o:(Kokkos::Impl::SharedAllocationRecord<void, void>* Kokkos::Impl::ViewMapping<Kokkos::ViewTraits<int*, Kokkos::HostSpace>, void>::allocate_shared<Kokkos::OpenMP, Kokkos::Impl::WithoutInitializing_t, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char>>, Kokkos::HostSpace>(Kokkos::Impl::ViewCtorProp<Kokkos::OpenMP, Kokkos::Impl::WithoutInitializing_t, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char>>, Kokkos::HostSpace> const&, Kokkos::LayoutRight const&, bool))
>>> the vtable symbol may be undefined because the class is missing its key function (see https://lld.llvm.org/missingkeyfunction)

Trying to see if the same thing happens with switching GCC build.