fix checkpointing with >2GB per process

[tjhei]

This fixes save/load of fixed and variable checkpointing where
individual ranks write more than 2GBs of data.
Part of #12873 and #12752
using facility added in #12964

[bangerth]

I suppose you have tested this with files this large?

[tjhei]

I suppose you have tested this with files this large?

Yes by using the supplied test and changing the bool to big=true. I only checked MPICH though.

[bangerth]

If I run this test with big==true and on two processes, it fails like so:

--------------------------------------------------------
An error occurred in line <1478> of file </raid/bangerth/p/deal.II/1/dealii/source/distributed/tria_base.cc> in function
    void dealii::parallel::DistributedTriangulationBase<dim, spacedim>::DataTransfer::save(unsigned int, unsigned int, const string&) const [with int dim = 3; int spacedim = 3; std::__cxx11::string = std::__cxx11::basic_string<char>]
The violated condition was: 
    ierr == MPI_SUCCESS
Additional information: 
    deal.II encountered an error while calling an MPI function.
    The description of the error provided by MPI is "MPI_ERR_ARG: invalid
    argument of some other kind".
    The numerical value of the original error code is 13.
--------------------------------------------------------

This is with the older OpenMPI version. (The comment you added makes me scared to run this on my laptop.) Is this to be expected?