ML: add option to replace MPI communicator

[jhux2]

@rhoope has asked whether it would be possible to update ML's preconditioner ctors to allow a user to pass in an MPI communicator that would replace the default (MPI_COMM_WORLD).

@csiefer2 @trilinos/ml

[csiefer2]

I was under the impression that (barring error cases) ML used whatever communicator the Epetra object you handed it has.

[jhux2]

That's what I see, as well. However, @rhoope is going to submit a patch, so we'll have a chance to see what he's proposing.

[csiefer2]

Reviewed changes to the C interface. They look good.

[rhoope]

I think this has been resolved as needed by Alegra and can be closed. I don't have permission to do so.