Testing: Make hdf5 test able to run in parallel and guard BP engine tests

Remove meaningless code(AddTransport which is not implemented yet) in
HDF5 test as well.

In this PR, I extend HDF5 native reader so that now hdf5 tests can read data
by selection(so called hyperslab selection in hdf5).

CDash labeling

CI improvements

Added mpi4py to build image and removed unwanted ohpc numpy package.

Use github api to retrieve real branch name, use that along with other
information to generate a meaningful build name.  To keep this from
happening too often, we only attempt to retrieve the branch name once
for each job during the first step, then communicate that name to
subsequent steps via an environment variable.  This required that we
stop using --login option when running the job steps from circleci,
as the lmod environment setup uses the same variable (BASH_ENV) to
store the path to the profile setup file that CircleCI sources for
each "run" in the job steps.

Use github api and the robot user oauth key (from CircleCI env) to post
an extra status message on each tested commit pointing to the CDash results.
By also adding an "update" step to the default build/test job and using
CTEST_UPDATE_VERSION_ONLY=TRUE, we can generate a deep link to the CDash
results specific to the appropriate commit.

Make Interface and HDF5 engine tests able to run in parallel

Add a variety of "shields" and badges to the readme

Make all Bp engine tests able to run in parallel

This PR also add testing for bp file name with path.

Since collective meta generation has not landed yet, I have to use
same attribute for all processes. Ideally attribute per process
should be unique to each other for testing purpose.

Finish image buildout

According to circle docs, if you include the path argument to "checkout":
if the path exists and is a git repo, no clone will be done, only fetch
from origin.  If path exists and is not a git repo, it is an error.