provider/lxd: fix off-by-1 LXD subnet calculation

[frobware]

The call to detectSubnet() was always adding 1 to the network address
that was discovered when parsing the output from `ip addr'. This has the
unfortunate effect of changing the subnet for the LXD bridge after it
has been recorded by Juju. This meant that further container <=>
apiserver communication would break as the original value was recorded
in the agent.conf file.

Fixes LP:#1569361

(Review request: http://reviews.vapour.ws/r/4538/)

[frobware]

$$merge$$

[jujubot]

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

[tych0]

I'm confused by this PR. Do we not then pick an existing subnet to configure lxdbr0 with? Why doesn't it break initially now?

[jameinel]

Right, the whole point of detectSubnet is to choose a subnet that isn't in use.
If we need to change it so that we detect the lxdbr0 and just return the one that already exists, that's fine, but the fix is to notice that we already have an address. detectSubnet should probably be renamed "findUnusedSubnet" or maybe split up?
Otherwise this is going to start creating conflicts when it finds an existing subnet.