Allow using native histograms for all components

[simonpasquier]

What is missing?

Since Prometheus v3.8.0, native histograms are considered a stable feature but there's no option to benefit from them in kube-prometheus.

Why do we need it?

In most cases, native histograms are more efficient and accurate than legacy histograms.

Environment

N/A

• kube-prometheus version:

N/A

Anything else we need to know?:

It also involves modifications to upstream repositories to update the alerting rules and dashboards.

Regarding the migration path:

1. Enable the following options
   a. Scraping of native histogram scraping
   b. Ingestion of classic histograms along with native histograms to keep existing monitoring operational (https://prometheus.io/docs/specs/native_histograms/#scraping-both-classic-and-native-histograms)
2. Consider if we should also convert classic histograms to native histograms. It might depend on the monitored components (for instance, when an alerting rule relies on a existing bucket boundary, it wouldn't be recommended).
3. Allow users to go "native histograms"-only mode at their own pace (typically after they've ingested native histograms for as long as their retention period).