This minimal snap downloads the kickstart.sh script during the pull
stage, executes the script without starting the Agent, and then ensures
it is properly packed.

This only awards the `network` and `network-bind` plugs to the Agent,
and the `network` plug for the claim script.

This includes many `disk.*` metrics, cpu and memory pressure,
`system.io`.

This includes `ip.*`, `ipv4.*`, `ipv6.*`, `wireless.*` and
`netfilter.conntrack_sockets`.

This is the only plug that provides wide read access to other processes'
properties, opening up proper metrics on file descriptors and limits.

However, I don't think this is proper and `snapd` should expose another
way to get this type of access, because it isn't necessarily beholden to
using Docker.

This enables `cgroups.plugin` to access unified cgroups files in
`/sys/fs/cgroup` and emit metrics for cgroups of various kinds,
including systemd services.

Note that systemd user slices (and their child slices) are disabled by
default by the plugin.