Change .container.restarts to be actionnable in a monitor

[xvello]

What does this PR do?

The kubernetes.containers.restarts metric is currently collected as a ever-increasing gauge, and tagged by container_id. This makes the value stay at 1 and change tags every time a container restarts inside a given pod:

This makes the metric very hard to use for alerting, while we would expect it to give an actual count of restarts per pod_name/kube_container_name. This PR does the following:

• change the collection type to a monotonic counter: the agent will compute the value delta and only send non-zero when a restart occurred
• move this gauge (and the other container state gauges) to orchestrator cardinality, meaning the timeseries will not churn when the container restart

Depends on #3413 , the target branch will be changed back to master after the first PR is merged.

Monitor example

Testing with 2 then 6 restarts of redis containers on a given host:

Before (gauge)

After

Review checklist (to be filled by reviewers)

• PR title must be written as a CHANGELOG entry (see why)
• Files changes must correspond to the primary purpose of the PR as described in the title (small unrelated changes should have their own PR)
• PR must have changelog/ and integration/ labels attached
• Feature or bugfix must have tests
• Git history must be clean
• If PR adds a configuration option, it must be added to the configuration file.

[stale]

This issue has been automatically marked as stale because it has not had activity in the last 30 days. Note that the issue will not be automatically closed, but this notification will remind us to investigate why there's been inactivity. Thank you for participating in the Datadog open source community.