Fix MPI tally reductions for counts larger than INT_MAX

[paulromano]

Description

This PR fixes silent tally corruption in MPI runs when a tally result buffer contains more than INT_MAX elements. The existing reductions passed size_t counts directly to MPI_Reduce, whose legacy C API accepts an int. Large counts could therefore be narrowed silently, leaving part of the tally unreduced. In my implementation, the large-count MPI_Reduce_c function from MPI-4 is used if available (checked via CMake capability detection). However, because MPI-4 support is not universal, it will fall back to correctly bounded MPI_Reduce chunks that are no more than INT_MAX elements.

Note that I was hesitant to add a test that allocates a > 2 GB buffer in CI, so at present there is no associated test.

Separately, while working on this, I ran into a compile-time error when building OpenMC with MPI against a LibMesh build that wasn't built against MPI; small fix for that is also included.

Checklist

• I have performed a self-review of my own code
• I have run clang-format (version 18) on any C++ source files (if applicable)
• I have followed the style guidelines for Python source files (if applicable)
• I have made corresponding changes to the documentation (if applicable)
• I have added tests that prove my fix is effective or that my feature works (if applicable)

[GuySten]

There are more places in the code using MPI_Reduce:

rg MPI_Reduce
src/mesh.cpp
1157:  MPI_Reduce(
1162:    MPI_Reduce(&outside_, outside, 1, MPI_C_BOOL, MPI_LOR, 0, mpi::intracomm);
1630:  MPI_Reduce(
1635:    MPI_Reduce(&outside_, outside, 1, MPI_C_BOOL, MPI_LOR, 0, mpi::intracomm);

src/tallies/tally.cpp
1053:      MPI_Reduce(values.data(), values_reduced.data(), values.size(),
1077:  MPI_Reduce(gt_values.data(), gt_values_reduced.data(), N_GLOBAL_TALLIES,
1090:  MPI_Reduce(&simulation::total_weight, &weight_reduced, 1, MPI_DOUBLE, MPI_SUM,

src/output.cpp
246:  MPI_Reduce(temp.data(), model::overlap_check_count.data(),

src/simulation.cpp
230:    MPI_Reduce(&simulation::simulation_tracks_completed, &total_tracks, 1,

src/cmfd_solver.cpp
120:  MPI_Reduce(
124:  MPI_Reduce(&outside_, outside, 1, MPI_C_BOOL, MPI_LOR, 0, mpi::intracomm);

src/state_point.cpp
907:  MPI_Reduce(gt.data(), gt_reduced.data(), gt.size(), MPI_DOUBLE, MPI_SUM, 0,
956:      MPI_Reduce(MPI_IN_PLACE, values.data(), values.size(), MPI_DOUBLE,
986:      MPI_Reduce(values.data(), nullptr, values.size(), MPI_DOUBLE, MPI_SUM, 0,

[paulromano]

I didn't touch the other reductions since they operate on small, fixed-size buffers just to limit the scope of this PR. I can update them though if you feel strongly.

[GuySten]

In some of the cases it is not clear to me that the reductions are over small size buffers.
Specifically the reductions in mesh.cpp and cmfd_solver.cpp.

[paulromano]

@GuySten Good point; I've updated a few of more variable-length reductions to be safe

[GuySten]

Looks good to me now.