Update the PMIx support to the PMIx 1.1.2 release.

[rhc54]

Fix intercomm_create with minor update to ompi/dpm and update of the test

[mellanox-github]

Test PASSed.
See http://bgate.mellanox.com/jenkins/job/gh-ompi-release-pr/1332/ for details.

[jsquyres]

@hppritcha Let's chat about this on Monday (Ralph and I chatted about this before he made this PR). It:

• updates PMIx to 1.1.2
• updates some infrastructure to handle the new things in PMIx 1.1.2
• fixes some memory leaks
• fixes MPI_INTERCOMM_MERGE

[hppritcha]

@jsquyres
@rhc54
all I can say is > 200 lines and 30 files changed just a few days before a RC2 speaks volumes about this project. we've got to get a better handle on release management. No wonder we talked about releasing 2.0 last June and here we are in 2016 in February and have yet to have a release.

[jsquyres]

@hppritcha Yes, we should talk about this one. It synchronizes an external project that doesn't conform to the same release schedule as Open MPI... hence it's working at a different pace. More importantly, it fixes some important bugs.

[hppritcha]

i'll give this PR a spin on one of the slurm/pmi1 clusters here.

[hppritcha]

👍
works for me with pmi1 on a slurm cluster.

[hppritcha]

@jsquyres this is ready to go.

[artpol84]

@jladd-mlnx @hppritcha @jsquyres @rhc54 to what changes in master this PR corresponds to?
Is this a cherry-picking or just a standalone set of commits that has nothing to do with master?

[artpol84]

Originally I thought that it is related to open-mpi/ompi@03eb1a8 but it doesn't seem like it is.
At least a) open-mpi/ompi@03eb1a8 has even more files changed; b) it doesn't introduce opal/errhandler directory that this commit introduces.

[artpol84]

The question is in the context of open-mpi/ompi#2129

[artpol84]

The question is in the context of open-mpi/ompi#2072

[jsquyres]

I believe the way @rhc54 does this is:

• Make a branch off the OMPI release branch
• Make a new component for the target PMIx release
• Un-tar the PMIx release tarball in that new component, trim/adjust it as relevant
• Cherry pick and/or adjust the rest of the code base as appropriate

I believe that Ralph has discussed this methodology in face-to-face meetings; that's why there's no corresponding master commits.