Allow using native histogram version of latency metrics in Mimir

[krajorama]

Is your feature request related to a problem? Please describe.

The classic histogram cortex_request_duration_seconds_(bucket|count|sum) now has a native histogram version called cortex_request_duration_seconds. Users should be able to drop the classic histogram series and just use the native histogram that has better quality and performance.

The solution should implement the above for all latency related metrics that Mimir has. Tracked in #5020 .

Describe the solution you'd like

Update recording rules and alerts to be able to use the new native histograms. It will be up to the end user to decide which ones to scrape.

Update dashboards to be able to show either classic or native histograms based on a dashboard variable.

Describe alternatives you've considered

We considered many alternatives:

• Simply use OR in PromQL between classic and native queries - but this can produce some unexpected results due to range vector selectors under rate working on very little data during migration.
• Duplicate the whole dashboard - makes it inconvenient to look at old data.
• Allow setting a timestamp to use as a guideline for PromQL when to query which data - can be inconvenient to set up.
• Try to show both classic and native at the same time - implementation became complex.
• Require users to keep both histograms inside their retention time - this is still recommended, but may be too expensive.
• Implement some kind of automatic detection / translation in Prometheus or Grafana - implementation became too complex.

Additional context

https://grafana.com/docs/mimir/v2.11.x/send/native-histograms/
https://grafana.com/docs/mimir/v2.11.x/visualize/native-histograms/