Remove MPI runtime requirement for ParMETIS test

[ocaisa]

The test run of ParMETIS is done without the use of MPIEXEC_EXECUTABLE and this can lead to false negatives with some MPI implementations. This change further trivialises the test code to avoid this issue (and, given the FIXME, is equivalent).

[poulson]

Please ignore the Travis failure: it was just a timeout. I am not sure of the utility of removing the MPI calls (MPI is required for ParMETIS), but adding an explicit ParMETIS call would be a much-needed addition.

[ocaisa]

Yes, MPI is required for ParMETIS but the existing test does not launch using MPIEXEC_EXECUTABLE (i.e., there is no real MPI environment). On our system with the latest Intel MPI the current test incorrectly results in

...and run output:
Abort(1094031) on node 0 (rank 0 in comm 0): Fatal error in PMPI_Init: Other MPI error, error stack:
MPIR_Init_thread(607)......: 
MPID_Init(793).............: 
MPIDI_NM_mpi_init_hook(667): OFI addrinfo() failed (ofi_init.h:667:MPIDI_NM_mpi_init_hook:No data available)

Return value: 143

which is purely because the compiled code was run directly rather than via MPIEXEC_EXECUTABLE. I don't know CMake well enough in order to suggest something but if you want to add the ParMETIS call (which I also don't know very well, I'm just installing Elemental for someone) then you would need to fix this.

[poulson]

I am no longer maintaining Elemental (I quit my job and am starting up new open source scientific computing projects). I prefer to know less about CMake than I do and have switched to Meson.

[ocaisa]

Ok, no problem ;)
Feel free to close this, we will just patch the release.
Good luck with the new projects!