perf: improve the memory usage of histogram

[xsbchen]

We have a use case where it generates a huge amount of histograms, each has more than 10 labels, we found the memory usage is quite high, under this scenario, we did an experiment and found that removing bucketExemplars field from histograms could save about 27% memory usage for each histogram instance, hope the maintainers could review and consider it

[SimenB]

Should we do this for the other metric types as well?

[xsbchen]

Should we do this for the other metric types as well?

I checked only Counter and Histogram support the enableExemplars option, and Counter already handled this case.

[SimenB]

@voltbit thoughts on this? Seems reasonable to me, but I have to admit I'm a bit unsure how this code works 😅

[SimenB]

From my read of it, this would work the same of exemplars are enabled, so should be good to go 👍

I'd like an extra pair of eyes on it, though

[voltbit]

Looks good to me. The data struct being there when exemplars are disabled was an overlook on my side I think.

[xsbchen]

@zbjornson @siimon master, plz help to review it if you are free, thanks.

[SimenB]

https://github.com/siimon/prom-client/releases/tag/v15.1.1