Fixed lp:1348663 - ensure NICs are brought down to trigger DHCPRELEASE

[dimitern]

Added a new "juju-clean-shutdown" init job/service to all machines at
initial boot (as an upstart job or systemd service, as needed). This job
ensures that on reboot, shutdown, or halt, "ifdown -a" is called to
bring down (and consequently issue a DHCPRELEASE packet) all machine
network interfaces.

A drive-by fix in the MAAS provider is included. I've discovered when
you bootstrap (or add) a vivid machine on MAAS the script creating
juju-br0 (calling ifdown PRIMARY_NIC ; ifup juju-br0) is somehow causing
a duplicate route (e.g. 10.14.0.0/24 proto kernel scope link dev et0 src
10.14.0.100 and 10.14.0.0/24 proto kernel scope link dev juju-br0 src
10.14.0.100) which effectively makes the bootstrap node inaccessible.
The drive-by fix just flushes any routes on the primary NIC before
bringing up juju-br0 - verified to work by live testing.

Live tested extensively on MAAS and EC2 using different combinations of
series (precise, trusty, vivid) and workloads in containers (LXC, KVM).
Each machine or container was manually stopped or rebooted and
DHCPRELEASE packets are observed as this happens on the MAAS DHCP server
side using dhcpdump.

(Review request: http://reviews.vapour.ws/r/1918/)

[dimitern]

$$fixes-1348663$$

[jujubot]

Status: merge request accepted. Url: http://juju-ci.vapour.ws:8080/job/github-merge-juju