kubectl run is being deprecated as a multi-purpose tool.
This PR replaces 'kubectl run' with 'kubectl create deployment'
in most places (except in the very first example, to reduce the
cognitive load; and when we really want a single-shot container).
It also updates the places where we use a 'run' label, since
'kubectl create deployment' uses the 'app' label instead.
NOTE: this hasn't gone through end-to-end testing yet.
ElasticSearch slowly uses up to 2GB of RAM.
Eventually, on instances provisioned with
only 4GB of RAM and without swap, if more
than one ElasticSearch pod end up on the
same instance, it will cause the instance
to slow down and ultimately crash. Instead,
we now use a tiny Go web server that shows
its environment in JSON. It still highlights
that multiple backends are serving requests
but without the memory usage issue.
This was discussed and agreed in #246. It will probably break a few
outstanding PRs as well as a few external links but it's for the
better good long term.
