Quick first feedback:

After some thinking, my initial hunch has solidified: The names are horrible. While it would be great to have the same name as in ViMe, I don't think I can live with the inconsistent naming here. I like your original idea timestamp_of_max_over_time much more, but it is a bit verbose. Maybe let's do something like ts_of_max_over_time, or max_over_time_ts?

The ViMe names introduce a prefix that is not separated by an underscore. It also doesn't make sense as tmax_over_time sounds like it returns the "time maximum" or "timestamp maximum" of something. But the returned timestamp is not a maximum.

Another general thing: These new functions have to be added to the frontend code, too. Please check other PRs that add new functions to see how that works, e.g. #13059

Really like this idea, would be super useful for certain peak value over time range queries/dashboard panels!
But yeah, maybe some neater naming :)

Looks pretty good. Found one bug with timestamp rounding.

@juliusv could you confirm that you are fine with this and double-check the frontend part? (or @Nexucis of course)

changes on UI side looks good to me.

Great stuff.

With @Nexucis UI check and given that this is behind the feature flag, I'll merge it now. If outrage ensues, we can just roll back. It's just an experiment.

Amazing, LGTM for the backend code. I can imagine those being super useful for detecting when certain events occurred e.g OOM or CPU spike or lost monitoring 👍