Session auto-restart options

Just like Docker, let's add a "restart" option when creating a session.Its value may be `none`, `unless-stopped`.

For `unless-stopped`, the session containers are automatically restarted when their agent hosts restart. Currently restarting the agent daemon itself preserves the containers but rebooting makes them to terminate.I don't think we need to add the `always` option like Docker.

Even we may simply pass those options directly to the Docker daemon API with some extra care on kernel-container synchronization upon agent startup (because there may be races between docker daemon's container auto-restarts and our agent startup procedures).

Q. For single-container sessions and single-node cluster sessions, the lifecycle of all belonging containers is naturally synchronized with the host. However, for multi-node cluster sessions, only a subset of containers in a session are affected by rebooting of specific agents. How could we handle such cases?

JIRA Issue: BA-268