Refactored mpirun to not exec a second executable

[gagarc]

Modified mpirun to no longer fork/exec prterun, but to instead instead call the function directly

[jsquyres]

Looks pretty good!

[hppritcha]

These changes look good, but we'll need to get the changes in prrte for adding prte_launch to the prte.h include file.

We will need for uofl-capstone-open-mpi/prrte#6 to be merged in then move the prrte sha in the capstone ompi repo forward to pick that change up before CI can pass for this PR.

[jsquyres]

@gagarc I think you need to rebase to the tip of capstone-devel. Github is saying you have conflicts with the submodule update.

You may find it easier to drop your prrte git submodule update, then rebase, then re-create your prrte git submodule update.

[jsquyres]

@hppritcha The mpi4py CI failure here is legit. It stems from the OMPI DPM looking for the prte executable and (correctly) not finding it. See https://github.com/uofl-capstone-open-mpi/ompi/blob/capstone-devel/ompi/dpm/dpm.c#L1978-L2016. I see that dpm.c:find_prte() is invoked from dpm.c:start_dvm() (i.e., https://github.com/uofl-capstone-open-mpi/ompi/blob/capstone-devel/ompi/dpm/dpm.c#L2047).

Do we need a PRTE library way to start a DVM?

[hppritcha]

oopss it looks like a change on the prrte side caused the prte command to vanish. we need that.

[hppritcha]

okay now just getting known comm create from group thing in mpi4py. rerunning.