Macro tweaks

[roystgnr]

In theory this PR is actually only dozens of lines of additions, hundreds of deletions, and thousands of moves (a bunch of code that had with-mpi and without-mpi versions now has a common implementation outside the ifdef LIBMESH_HAVE_MPI block), but my git version doesn't support --color-moved and I can't figure out how to make Github use it either.

In practice I did have a few rebase conflicts; hopefully Civet will verify that I didn't screw up any of the resolutions.

[jwpeterson]

There are reams of template error messages in the CIVET output, here is the first one for Communicator::min() FWIW.

In file included from ./include/libmesh/parallel.h:47:0,
                 from ./include/libmesh/libmesh.h:27,
                 from ../src/base/libmesh_common.C:20:
./include/libmesh/parallel_implementation.h:2103:13: error: prototype for ‘void libMesh::Parallel::Communicator::min(bool&) const’ does not match any in class ‘libMesh::Parallel::Communicator’
 inline void Communicator::min(bool & r) const
             ^~~~~~~~~~~~
In file included from ./include/libmesh/parallel.h:24:0,
                 from ./include/libmesh/libmesh.h:27,
                 from ../src/base/libmesh_common.C:20:
./include/libmesh/communicator.h:237:8: error: candidate is: template<class T> void libMesh::Parallel::Communicator::min(T&) const
   void min(T & r) const;
        ^~~

[roystgnr]

Thanks; I can reproduce the problems and I've got them mostly fixed. This PR got interrupted by my trip while I was 80% done with it and when I got back I'd forgotten that the other 20% included --disable-mpi testing...

[roystgnr]

Planning to merge this soon; it's only a minor improvement but moving all that code is going to cause no end of conflicts when anyone tries to play with moved sections of parallel_implementation.h while it's outstanding. Fortunately the new method in #1826 is surrounded by unmoved code so should be safe.

[jwpeterson]

The no-MPI test should already be covered by the --enable-complex case, but since there were so many changes I thought it would be good to test the no MPI and !LIBMESH_USE_COMPLEX_NUMBERS configuration...

[jwpeterson]

This is handy... there are several places in the examples where we ignore multiple variables in the same scope.

[jwpeterson]

Looks like a big improvement to having ifdefs all over the place.