Cleanup MPI calls

[ldowen]

Summary

This PR improves the interface to the MPI functionality.

• This PR is a refactoring and bugfix
• It does the following:
  • Refactors the allReduce and scan wrappers to use the Communicator::communicator() by default and puts both in a single file.
  • Refactors the Distributed source directory, having it provide the MPI wrappers even when MPI is not enabled; this prevents potential bugs from occurring when wrappers like Communicator are present but Communicator.hh is not available.
  • Replaces the MPI_SUM, MPI_MAX, etc with SPHERAL_OP_SUM, SPHERAL_OP_MAX etc

---

ToDo :

• Annotate RELEASE_NOTES.md with notable changes.
• Create LLNLSpheral PR pointing at this branch. (PR#88)
• LLNLSpheral PR has passed all tests.

[mdavis36]

What do you guys think about using something like SPHERAL_OP_MIN, SPHERAL_OP_MAX etc. Instead of SPHERAL_MPI_MIN, SPHERAL_MPI_MAX. It feels like those macros suggest an MPI call instead of just defining an execution context agnostic operator.

[ldowen]

Yeah I see what you mean. That works for me. I will change it now.