[processor/cumulativetodelta] Convert cumulative Histograms to delta temporality

[mistodon]

Is your feature request related to a problem? Please describe.
It seems that cumulative-to-delta at the moment only processes Sums, and just skips over other data types. This is a problem for us, as we have Histograms in our pipeline with cumulative temporality that we want to convert.

Describe the solution you'd like
Rather than bypass Histogram metrics, have the cumulative-to-delta processor convert their data points in a similar way to sums.

Describe alternatives you've considered
None, other than continuing to ignore non-Sum metrics.

[TylerHelmuth]

Yes I would love for this capability to exist. The reason it hasn't been implemented yet is that this processor will hopefully be replaced some day by a stateful version of the transform processor. I've been hesitant to make changes to it because I'd like to deprecate it; any functionality added is functionality that has to be moved over to the new processor.

In this case tho, I feel like this is an important enough feature (and no one has started on a stateful transform processor yet 😭) that I think we could add it now to the cumulativetodelta processor. Is it something you plan to work on?

[mistodon]

Thanks for the quick update! That makes a lot of sense - but given that the new processor isn't looking like it'll be ready super soon, I agree it's worth adding now.

I have started looking at implementing this. I'm not extremely familiar with the codebase or anything, but once I have something, I can open a draft PR for feedback if that sounds good?

[TylerHelmuth]

Sounds great

[mistodon]

There's definitely some things that could be refactored in the above PR, but I want to make sure people are happy with the general approach before I do that.

[TylerHelmuth]

I think your approach is good. I would say to switch to using a struct instead of a list for the 3 histogram identities, that way we don't have to use any index assumptions.

Ping me when you are ready for a full review. Thanks for doing this!