Add dockerfile mpi

[jcurtis2]

Adding a dockerfile for testing of PartMC compiled with MPI.

[jcurtis2]

@mwest1066 if you could take a look at this warning in the Dockerfile.mpi build that is turning into an error in the build process with our compiler flags, I don't quite see why gfortran complains about it (after working through all the logic of that subroutine). It complains about the following lines:

https://github.com/compdyn/partmc/blob/0e8cf41bf2d783005f34998789862ce545d5f6c1/src/bin_grid.F90#L370-L371

saying "Error: Impure function 'pmc_mpi_allequal_real' at (1) might not be evaluated [-Werror=function-elimination".

There's a few ways to avoid this problem but I wasn't sure if I was missing something (we can split the if statement into two) or we can get the logicals separately and compare:

    logical :: same_bin_min, same_bin_max

    same_bin_min = pmc_mpi_allequal_real(val%edges(1))
    same_bin_max = pmc_mpi_allequal_real(val%edges(bin_grid_size(val)))
    if (bin_min .and. bin_max) then
       pmc_mpi_allequal_bin_grid = .true.
    else
       pmc_mpi_allequal_bin_grid = .false.
    end if

[mwest1066]

I think your solution with same_bin_min, same_bin_max is good. Let's use that.

The reason that it's warning is:

1. If pmc_mpi_allequal_real(val%edges(1)) is false
2. Then Fortran will short-circuit the evaluation of the .and.
3. Thus it will not even compute pmc_mpi_allequal_real(val%edges(bin_grid_size(val)))
4. But pmc_mpi_allequal_real() is "impure" (that is, it is not marked as "pure")
5. Therefore the compiler can't be sure that the possibly-missing second call to pmc_mpi_allequal_real() wasn't having an important side-effect
6. Thus we get a warning/error

Your fix with same_bin_min, same_bin_max solves this problem because it forces the code to always call pmc_mpi_allequal_real() twice, avoiding the short-circuit .and.. The nested-if solution can short-circuit (if the outside if is false then it won't try the inner if) but this is ok because it's explicitly in the code, so the compiler won't warn.

[jcurtis2]

https://docs.github.com/en/actions/using-github-hosted-runners/about-github-hosted-runners#supported-runners-and-hardware-resources

Github hardware is a currently a 2 core CPU so the MPI code was adjusted to 2 cores for now so tests pass.