[Aram164]

Refs #198

Port PDS Q2 from polars-benchmark to a skrub DAG and check Stratum against the original pandas query on the same Arrow-backed fixtures, following the existing Q1 pipeline test in tests/application/pds/test_q1_pipeline.py.

What it checks

• Five-table joins, the three filters, the grouped minimum, and the join-back on its shared filtered input, on both pandas and Polars.
• Constant folding of var1/var2/var3, physical operator selection, and the pandas_query selection flag.
• Projection, mixed-direction sorting, and head(100), exercised with 105 qualifying suppliers so the limit actually drops rows.

The fixtures include tied minimum-cost suppliers and a cheaper supplier outside the selected region, plus wrong-size, wrong-type, and Asia-only parts so each filter is isolated. The prefix comparison normalizes row order, the full-query comparisons check the final sorted order.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Flag	Coverage Δ
unittests	92.78% <ø> (+0.04%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.
see 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.

[e-strauss]

Thanks for the Q2 port, @Aram164.

Before merging I made a two minor changes:

• Column selection: build_q2 now uses jn2[[...]] instead of jn2.skb.select([...]). Plain indexing is the more natural way to pick a fixed list of columns, and it's closer to the .loc[:, cols] in the original query. skb.select is mainly useful for skrub selectors. Indexing lowers to a ColumnProjectionOp instead of a ColumnSelectorOp, so test_q2_folds_the_sort_and_limit now checks that op, along with its .key and the PolarsColumnProjectionOp it binds to under greedy.
• Greedy parity: test_q2_matches_reference now uses the polars fixture, so the column-order check against the reference runs with the greedy selector too.