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.