Improve Prometheus and Grafana integration

[Kevin W Monroe]

Currently, conjure-up scrapes k8s metrics via prometheus with this:

https://github.com/conjure-up/spells/blob/master/charmed-kubernetes/addons/prometheus/steps/01_install-prometheus/prometheus-scrape-k8s.yaml

This could use improvement:

1 - better auth (token or other service account vs admin creds)
2 - scrape via the kubeapi-lb instead of hitting endpoints directly
3 - create prometheus scrape job for k8s automatically via a relation
4 - create grafana dashboard for k8s automatically via a relation

#2 is the reason this bug is filed here. If we find everything we need via the kubeapi-lb proxy, we do not need to expose k8s-[master|worker] units.

[Kevin W Monroe (kwmonroe)]

See https://github.com/juju-solutions/bundle-canonical-kubernetes/issues/687#issuecomment-438202521 for context on when certain endpoints need to be exposed.

[Cory Johns (johnsca)]

This will be fixed when https://github.com/charmed-kubernetes/charm-kubernetes-master/pull/45 and all of its dependencies are reviewed, merged, and released.

[Cory Johns (johnsca)]

In addition to #1 and #2, the linked PR(s) significantly simplifies the installation story by using relations to manage the scraper jobs and dashboards. See updated instructions in https://github.com/charmed-kubernetes/kubernetes-docs/pull/247