Set read/write tracking flags when passing debug array to MPI.Buffer()

If a shared-memory array (or part of one) is passed to `MPI.Buffer()` it
will be communicated. This probably implies it is read on at least one
of the distributed-MPI processes, and read on others. To check for
errors as conservatively as possible (I think, but not sure if this is
really true), set the `is_read`, `is_written` and `is_initialized` and
`accessed` flags of an `MPIDebugSharedArray` when it is passed to
`MPI.Buffer()`, so that a shared-memory synchronization is required
before the array is used. This should be a sensible thing to do because:
distributed-MPI communication is done by the root process of each
shared-memory block, so any array that is touched by distributed-memory
communication requires a synchronization before processes in the
shared-memory block (apart from the root process) are allowed to use it.

moment_kinetics/src/communication.jl

@@ -513,6 +513,10 @@ end
 
     import MPI: Buffer
     function Buffer(A::DebugMPISharedArray)
+        A.is_initialized .= 1
+        A.is_read .= true
+        A.is_written .= true
+        A.accessed[] = true
         return Buffer(A.data)
     end