Prior to this, if a flagged cell was on the processor boundary, and the neighbor cell on the other processor wasn't flagged, the buffers from FlagBufferZones wouldn't be built out. This PR changes two things.
Grid_FlagBufferZones is only called once. Now the loop over NumberOfBufferZones occurs within G_FBZ.
We buffer around all cells except the last outer face, then at the end of the NumberOfBufferZone loop, we clear out the ghost zone cells.