Describe the bug
When running a closed case of TE with the lastest versions, GEOS crashes (SegFault).

To Reproduce
I cannot join a case, but I'm posting a PR that will resolve this crash.
It happens because, if I understood correctly, in CompositionalMultiphaseFVM::scalingForSystemSolution, we assume that all MPI ranks will contains some information (pressure, density, temperature) for all the sub-regions, but in our case an MPI rank only has a non-simulated region.
The function scalingForSystemSolution() has an unsafe derefencing: we should not do *std::max_element() without checking if the list has any element at all.

Expected behavior
The function scalingForSystemSolution() should not dereference without checking if there is any information in the pressure/delta-density/temperature sets.

Platform (please complete the following information):

• Machine: Pangea4
• Compiler: gcc 12.2.1, OpenMPI
• Version: develop since the PR feat: Nonlinear Max Change Reporting to include cell location #3327 was merged.

Additional info
I'm posting a quickfix PR that will expose in MpiWrapper methods to safely reduce pairs or pair sets over ranks. I'm also adding a unit test to check the consistency of the solution, to be sure that the reduction will remain stable.