Ensure MPI_ABORT is called on errors

[rplzzz]

See details here:

https://mpi4py.readthedocs.io/en/stable/mpi4py.run.html

[rplzzz]

Confirmed: this bug will cause the system to hang if one of the components throws an exception. During the development of the MP branch, at one point I was iterating over the dictionary of outstanding tasks and removing the ones that had completed. This should have raised an exception and crashed the system. For example:

d = {1:2, 3:4, 5:6}
for k in d:
    if k==3:
        print(d.pop(k))

When you run this you get:

4
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
RuntimeError: dictionary changed size during iteration

When it happened in the MP code, however, the result was for the whole system to hang, with no output to stdout.

[rplzzz]

Note also that the fix proposed in the mpi4py docs doesn't require any change to our code, just a change to how we run it (using python -m mpi4py). I think we should be able to just add this to the shebang line of cassandra_main.py. We'd also want to test it by adding an option to the dummy component to throw an exception.

[rplzzz]

Well, crud. The fix described above only works in mpi4py 2.1.0 and above, and PIC has version 2.0.0 installed. Instead of trying to get them to upgrade, I'm going to see if I can just fix the problem by wrapping our main in a try block.