[sdk-metrics] Support exemplars when using exponential histograms

[CodeBlanch]

Changes

• Adds support for exemplars when using exponential histograms.

• Improves test coverage for the permutations of the MetricPoint.UpdateWithExemplar methods (double & long):

  

Merge requirement checklist

• CONTRIBUTING guidelines followed (license requirements, nullable enabled, static analysis, etc.)
• Unit tests added/updated
• Appropriate CHANGELOG.md files updated for non-trivial changes

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 84.27%. Comparing base (6250307) to head (5f97f42).
Report is 102 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #5396      +/-   ##
==========================================
+ Coverage   83.38%   84.27%   +0.88%     
==========================================
  Files         297      284      -13     
  Lines       12531    12080     -451     
==========================================
- Hits        10449    10180     -269     
+ Misses       2082     1900     -182     

Flag	Coverage Δ
unittests	?
unittests-Solution-Experimental	84.27% <100.00%> (?)
unittests-Solution-Stable	84.34% <100.00%> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.

Files	Coverage Δ
src/OpenTelemetry/Metrics/MetricPoint.cs	91.71% <100.00%> (+23.24%)	⬆️

... and 52 files with indirect coverage changes

[cijothomas]

i am not able to see the need for custom sized (3) for this instrument, and not others? Are we validating them separately?

[CodeBlanch]

Good catch. I removed the view from this test. Probably got there copy/pasting 🤣 Observables are an interesting case for exemplars. We only observe them during collect so there will only ever be at most 1 exemplar exported regardless of reservoir size. If user is doing TraceBasedExemplarFilter then they likely will never see an exemplar for observables because they are collected outside of all traces unless you go and force a trace in the observe callback which feels very strange. Might need to do something about that in SDK/spec?

[cijothomas]

yes. I want to see if spec would change the defaults, so that exemplars are opt-in. specifically, users must opt-in to exemplars for each instrument. (The key reasoning being - most users just need exemplar for one key instrument, typically the one measuring latency. The current model enables Exemplars for every instrument, including Observables as well, leading to wasted cpu/memory etc.)

[cijothomas]

LGTM

[utpilla]

@CodeBlanch Do you want to update the OTLP Exporter as well to export exemplars for Exponential Histograms?

[CodeBlanch]

@utpilla

Do you want to update the OTLP Exporter as well to export exemplars for Exponential Histograms?

Going to be my next PR!