Various native histogram spec updates

[beorn7]

See commit descriptions for details.

Let's only merge this once we are positive about how and when to tackle the related issues:

• PromQL (histograms): counter reset hint (AKA gauge vs. counter) is inconsistent when aggregating native histograms prometheus#17308 (I plan to create a PR today or tomorrow).
• promql: NHCB operations should support different bucket layouts prometheus#17255 (which already has a PR under review)

FYI:

• @linasm (Please check if I caught the implications of NHCB reconciliations correctly.)
• @crush-on-anechka (WRT counter reset hint implications – but as said, I plan to create a PR and will cc you on it.)

[linasm]

LGTM for the parts related to NHCB custom values (validation, reconciling).
I'm not familiar enough with counter reset hints, someone else should review that part.

[linasm]

Related question: it is not very clear from the spec, whether -Inf is a valid (first, obviously) custom bound value. Currently, Validate() would reject that, likely accidentally (no tests are failing if I change that check).

Also the spec should state that -Inf is inclusive first lower bound. Otherwise it is not clear where observations with -Inf value would be counted. For the sake of symmetry - observations with +Inf value are counted in the last bucket.

[beorn7]

Right, we should make the handling of -Inf explicit (as we do for the standard exponential schemas).

[beorn7]

I added a commit for that.

WRT -Inf as the 1st custom value: I think it should be allowed. That's implicit from the spec. I don't think we have to make it explicit in the spec. Just fix the Validate method. (Will create a PR momentarily.)

[beorn7]

See prometheus/prometheus#17320