[METRICS SDK] Enforce a runtime minimum scale for Base2ExponentialHistogramAggregation

[ThomsonTan]

Fixes #4325

Changes

This change enforces a runtime minimum scale of -10 for base2 exponential histograms, guarantees sufficient bucket capacity at that floor to preserve recorded counts, centralizes scale reduction across record, merge, and diff paths, guards the indexer against oversized shifts, and adds regression tests plus changelog documentation for the new behavior.

For significant contributions please make sure you have completed the following items:

• CHANGELOG.md updated for non-trivial changes
• Unit tests have been added
• Changes in public API reviewed

[codecov]

Codecov Report

❌ Patch coverage is 89.88764% with 9 lines in your changes missing coverage. Please review.
✅ Project coverage is 82.48%. Comparing base (3893576) to head (a5af8bd).

Files with missing lines	Patch %	Lines
...egation/base2_exponential_histogram_aggregation.cc	89.29%	9 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #4353      +/-   ##
==========================================
- Coverage   82.49%   82.48%   -0.00%     
==========================================
  Files         502      502              
  Lines       19899    19934      +35     
==========================================
+ Hits        16414    16441      +27     
- Misses       3485     3493       +8     

Files with missing lines	Coverage Δ
...metry/sdk/metrics/aggregation/aggregation_config.h	100.00% <ø> (ø)
...aggregation/base2_exponential_histogram_indexer.cc	100.00% <100.00%> (ø)
...egation/base2_exponential_histogram_aggregation.cc	94.78% <89.29%> (-1.84%)	⬇️

... and 1 file with indirect coverage changes

🚀 New features to boost your workflow:

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

[lalitb]

This can create 3 buckets when the user configured max_size = 2. Since max_size is the maximum number of buckets, can we use -11 as the runtime floor instead? At -11, the full double range fits in 2 buckets, so we do not need this exception.

[ThomsonTan]

Good catch, thanks. Changed the runtime floor to -11 and dropped kMinBucketsAtFloor, so max_size is now a hard cap. At -11 every finite double maps to an index in [-1, 0], and the max_size = 2 repro from #4325 now ends at scale -11 with a two-bucket span and capacity 2.