When device protocol goes down, interfaces should be flushed asynchronously (in the same way like routes from protocols are flushed), when protocol goes to DOWN/HUNGRY. This fixes the problem with static routes staying in kernel routing table after BIRD shutdown.