Use O(log n/eps) budget for CPU distributed sketch

[RAMitchell]

Summary

Use count-aware O(log n / eps) sizing for CPU distributed quantile sketch merge/prune.

This change makes WQuantileSketch track the number of represented elements, serializes those per-feature counts through the CPU distributed sketch payload, and recomputes SketchSummaryBudget(...) after merge using the summed counts from both sides.

Testing

• cmake -S <worktree> -B <worktree>/build-cpu -DUSE_CUDA=OFF -DGOOGLE_TEST=ON -DUSE_DMLC_GTEST=ON
• cmake --build <worktree>/build-cpu --target testxgboost -j35
• <worktree>/build-cpu/testxgboost --gtest_filter='Quantile.*'

[Copilot]

Pull request overview

Updates CPU distributed quantile sketching to size merge/prune budgets using a count-aware O(log n / eps) summary budget, by tracking and propagating per-feature represented element counts through the allreduce payload.

Changes:

• Add WQuantileSketch::NumElements() and track represented element counts during Push/PushSorted.
• Extend the distributed sketch allreduce payload to serialize/merge per-feature element counts and recompute SketchSummaryBudget(...) after merges.
• Add unit tests to validate element-count tracking behavior.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

File	Description
tests/cpp/common/test_quantile.cc	Adds tests for the new NumElements() tracking API.
src/common/quantile.h	Introduces num_elements_ tracking and a NumElements() accessor in WQuantileSketch.
src/common/quantile.cc	Updates distributed sketch serialization/allreduce merge logic to propagate element counts and recompute per-feature budgets.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.