Fix order preservation in grouped-window evaluation

[rjzamora]

Description

Follow-up to #23466

That PR introduced support for input-order-sensitive over(...) expressions without an explicit order_by argument. However, there is a bug: We do not properly sort by the original local row-position (only the rank and chunk index). Since the hash-partitioning of the forward shuffle is not technically "stable", we need to include this local row position in the forward sort operation as well.

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: 609c6ab1-85b7-4a30-a732-07846bb3f2b8

📥 Commits

Reviewing files that changed from the base of the PR and between 12ac15c and c6a6c1d.

📒 Files selected for processing (2)

• python/cudf_polars/cudf_polars/streaming/actor_graph/over.py
• python/cudf_polars/tests/streaming/test_spmd.py

---

📝 Walkthrough

Summary by CodeRabbit

• Bug Fixes

  • Improved deterministic ordering behavior in window operations to ensure consistent within-chunk input-order preservation.

• Tests

  • Added streaming test case for partitioned data sources with sorting and cumulative aggregation operations.

Walkthrough

The stamped-row sort now preserves input order within origin chunks by including row position. A single-rank streaming test validates cumulative grouped window results from a partitioned Parquet scan.

Changes

Window ordering

Layer / File(s)	Summary
Add origin row ordering python/cudf_polars/cudf_polars/streaming/actor_graph/over.py	The stamped-row sort adds origin row position as an ascending key after origin rank and chunk index.
Validate partitioned cumulative windows python/cudf_polars/tests/streaming/test_spmd.py	A single-rank streaming test scans partitioned Parquet data, sorts by x, computes cumulative sums over g, and validates the result.

Estimated code review effort: 2 (Simple) | ~10 minutes

Possibly related PRs

• rapidsai/cudf#23306: Updates the same window evaluation path with stamped-row input-order sorting.
• PR 23466: Adds the stamped-row input-order sorting that this change extends with origin row-position ordering.

Suggested labels: improvement

Suggested reviewers: mroeschke, nirandaperera, tomaugspurger

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title directly addresses the main change: fixing order preservation in grouped-window evaluation, which matches the core fix described in the PR objectives.
Description check	✅ Passed	The description explains the bug fix context, references a related PR, and details the specific problem with hash-partitioning stability and the required sort operation changes.
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.}

[rjzamora]

/merge