feat: use time-unbiased algorithm for histogram reservoir

[dashpole]

Fixes #8236

Factors out algorithm L into a nextTracker, and uses it for both the histogram reservoir and the fixed-size reservoir.

Benchmarks:

goos: linux
goarch: amd64
pkg: go.opentelemetry.io/otel/sdk/metric/exemplar
cpu: AMD EPYC 7B12
                           │  main.txt   │         new_embedded_2.txt          │
                           │   sec/op    │    sec/op     vs base               │
FixedSizeReservoirOffer-24   186.0n ± 4%   173.5n ±  3%   -6.75% (p=0.002 n=6)
HistogramReservoirOffer-24   46.39n ± 4%   20.35n ± 11%  -56.14% (p=0.002 n=6)
geomean                      92.91n        59.42n        -36.04%

Gemini helped write this.

I'll wait to mark this ready for review until after #8309

[codecov]

Codecov Report

❌ Patch coverage is 97.18310% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 82.9%. Comparing base (4086701) to head (c852a82).

Files with missing lines	Patch %	Lines
sdk/metric/exemplar/next_tracker.go	93.1%	1 Missing and 1 partial ⚠️

Additional details and impacted files

@@          Coverage Diff          @@
##            main   #8306   +/-   ##
=====================================
  Coverage   82.9%   82.9%           
=====================================
  Files        314     315    +1     
  Lines      24985   25005   +20     
=====================================
+ Hits       20730   20749   +19     
  Misses      3882    3882           
- Partials     373     374    +1     

Files with missing lines	Coverage Δ
sdk/metric/exemplar/fixed_size_reservoir.go	100.0% <100.0%> (+5.2%)	⬆️
sdk/metric/exemplar/histogram_reservoir.go	94.8% <100.0%> (+3.9%)	⬆️
sdk/metric/exemplar/storage.go	100.0% <100.0%> (ø)
sdk/metric/exemplar/next_tracker.go	93.1% <93.1%> (ø)

... and 3 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

Pull request overview

This PR updates the SDK metric exemplar reservoirs to use a time-unbiased reservoir sampling approach (Algorithm L) for histogram buckets, while factoring the sampling logic into a shared nextTracker used by both histogram and fixed-size reservoirs.

Changes:

• Introduces nextTracker implementing Algorithm L and reuses it across reservoirs.
• Updates FixedSizeReservoir and HistogramReservoir to use nextTracker-based sampling and refactors storage/locking accordingly.
• Adds a statistical test asserting time-unbiased behavior for the histogram reservoir.

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
sdk/metric/exemplar/storage.go	Removes the old shared storage wrapper and moves storage logic onto measurement.
sdk/metric/exemplar/next_tracker.go	Adds the shared Algorithm L sampling tracker.
sdk/metric/exemplar/histogram_reservoir.go	Switches histogram reservoir sampling from “last seen” to Algorithm L per bucket.
sdk/metric/exemplar/histogram_reservoir_test.go	Adds a statistical test to validate time-unbiased sampling behavior.
sdk/metric/exemplar/fixed_size_reservoir.go	Refactors fixed-size reservoir to use nextTracker and inline storage.
sdk/metric/exemplar/fixed_size_reservoir_test.go	Updates tests to reflect the new internal storage layout.
sdk/metric/exemplar/benchmark_test.go	Updates benchmark reset logic to use the new tracker.