Follow up from PR #3346 review [comment](https://github.com/open-telemetry/opentelemetry-cpp/pull/3346#discussion_r2043260347) 1. Base2 Exponential Histogram Aggregation downscaling in-place. 2. Aggregation benchmarks for throughput when multiple threads are involved. Something along the lines of what’s done in [measurements_benchmark.cc](https://github.com/open-telemetry/opentelemetry-cpp/blob/main/sdk/test/metrics/measurements_benchmark.cc)