Fluid approximations for a processor sharing queue