Workload analysis of a two-queue fluid polling model