Add instructions on running PDS-H for pandas

[Matt711]

Description

Documentation for humans and agents to refer to for reproducing our benchmarks.

Also removes CPU-benchmarking on CPU-only machines which we dont support right now.

Contributes to #17640

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 40840de6-32aa-4232-aa5a-a8c2f0312cef

📥 Commits

Reviewing files that changed from the base of the PR and between 1984137 and 3a2f4fd.

📒 Files selected for processing (1)

• docs/cudf/source/cudf_pandas/benchmarks.md

✅ Files skipped from review due to trivial changes (1)

• docs/cudf/source/cudf_pandas/benchmarks.md

---

📝 Walkthrough

Summary by CodeRabbit

• Documentation
  • Added a new step-by-step guide for running PDS-H benchmark results with cudf.pandas, including setup, data generation, data preparation, benchmark execution, and output format details.
  • Simplified the cudf-polars benchmark instructions by removing outdated GPU/CPU-specific installation guidance and streamlining the setup flow.

Walkthrough

This PR adds a new "PDS-H (TPC-H variant)" documentation section to the cudf.pandas benchmarks guide, covering installation, dataset generation with Decimal/date type conversion, CPU/GPU benchmark execution commands, and JSONL results format. It also simplifies the cudf-polars benchmarks documentation by removing GPU-specific preface text and the CPU-only source installation section.

Changes

Benchmark Documentation Updates

Layer / File(s)	Summary
cudf.pandas PDS-H benchmark guide docs/cudf/source/cudf_pandas/benchmarks.md	Adds a new documentation section with setup instructions for installing cudf nightly wheels and tpchgen-cli, dataset generation steps including a PyArrow script to cast Decimal to float64 and dates to millisecond timestamps, CPU/GPU benchmark run commands via python -m cudf.pandas._benchmarks.pdsh, and documentation of the JSONL results output format.
cudf-polars installation instructions simplified docs/cudf/source/cudf_polars/benchmarks.md	Removes the GPU-specific preface sentence and deletes the entire CPU-only (no CUDA) source installation section, streamlining setup to proceed directly to tpchgen-cli installation.

Estimated code review effort: 1 (Trivial) | ~5 minutes

Possibly related PRs

• rapidsai/cudf#22619: Both PRs update the cuDF Polars benchmark documentation related to PDS-H/PDS benchmark reporting.
• rapidsai/cudf#23025: Both PRs modify the same docs/cudf/source/cudf_polars/benchmarks.md setup/run instructions.

Suggested labels: `improvement`

Suggested reviewers: `rjzamora`, `TomAugspurger`, `mroeschke`

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly summarizes the main change: new instructions for running PDS-H benchmarks for pandas.
Description check	✅ Passed	The description is directly related to the documentation update and the removal of unsupported CPU-only benchmarking.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 2

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@docs/cudf/source/cudf_pandas/benchmarks.md`:
- Around line 125-139: The benchmark setup steps are missing the `pyarrow`
prerequisite, so the later casting workflow will fail in a clean environment.
Update the prerequisites in the `cudf_pandas` benchmark setup section to install
`pyarrow` alongside `cudf` and `tpchgen-cli`, and make sure the instructions
around the benchmark dataset generation mention this dependency so the
`pyarrow`/`pyarrow.parquet` imports succeed.
- Around line 173-179: The table conversion loop in the benchmarks flow can
silently skip a table when `parts` is empty, so update the logic around the `for
table in tables` block to fail fast instead of doing nothing. In the same path
that builds `table_path` and `parts`, add an explicit check for the empty case
and raise an error with a clear message when no parquet files are found for a
table, so unexpected layouts don’t go unnoticed.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: cd5524d1-beb3-4d42-abd9-2ad428cc9405

📥 Commits

Reviewing files that changed from the base of the PR and between a82686a and 1cb6d5c.

📒 Files selected for processing (3)

• docs/cudf/source/cudf_pandas/benchmarks.md
• docs/cudf/source/cudf_polars/benchmarks.md
• python/cudf/cudf/pandas/_benchmarks/utils.py

[Matt711]

/merge