Dump only on processor 0

- Some combination of linux and Python 3.8 fails to close /dev/null
  on other processors.
- Writing on only one file deadlocks in
  `fipy.meshes.representations.gridRepresentation._GridRepresentation._test`:
  https://app.circleci.com/jobs/github/usnistgov/fipy/1226
  https://app.circleci.com/jobs/github/usnistgov/fipy/1225
  https://app.circleci.com/jobs/github/usnistgov/fipy/1217
  https://travis-ci.org/usnistgov/fipy/jobs/627999300
- Stipulate binary

  Avoid doubling memory with `dumps()`,
  but prevent `/dev/null` closure problem

fipy/tools/dump.py

@@ -50,9 +50,9 @@ def write(data, filename = None, extension = '', communicator=parallelComm):
             (f, _filename) =  tempfile.mkstemp(extension)
         else:
             (f, _filename) = (None, filename)
-        fileStream = gzip.GzipFile(filename = _filename, mode = 'w', fileobj = None)
+        fileStream = gzip.GzipFile(filename=_filename, mode='wb', fileobj=None)
     else:
-        fileStream = open(os.devnull, mode='w')
+        fileStream = open(os.devnull, mode='wb')
         (f, _filename) = (None, os.devnull)
 
     pickle.dump(data, fileStream, 0)
@@ -78,7 +78,7 @@ def read(filename, fileobject=None, communicator=parallelComm, mesh_unmangle=Fal
         Whether to correct improper pickling of non-uniform meshes (ticket:243)
     """
     if communicator.procID == 0:
-        fileStream = gzip.GzipFile(filename = filename, mode = 'r', fileobj = None)
+        fileStream = gzip.GzipFile(filename=filename, mode='r', fileobj=None)
         data = fileStream.read()
         fileStream.close()
         if fileobject is not None: