fix: preserve aggregate filter pushdown order

[discord9]

Which issue does this PR close?

• ClosesAggregateExec filter pushdown can reorder parent filter results #22925.

Rationale for this change

FilterPushdown maps each child pushdown result back to its original parent
filter by position. AggregateExec::gather_filters_for_pushdown previously
split parent filters into safe and unsafe buckets and then concatenated the
results, which changed their order.

For example, in cnt@2 = 1 AND b@1 = bar, the grouping-column predicate on
b can cross the aggregate, while the predicate on aggregate output cnt
must remain above it. Reordering the returned results could make the optimizer
associate the pushed-down b result with the cnt predicate and incorrectly
remove the latter.

Aggregate pushdown also needs to account for empty-input semantics. A global
aggregate, or a grouping-sets aggregate containing (), can emit a row even
when its input is empty. Moving any parent predicate below such an aggregate —
including a column-free predicate such as false — can therefore change the
result.

What changes are included in this PR?

• Preserve parent-filter order by constructing each child description once
  with ChildFilterDescription::from_child_with_allowed_indices.
• Allow only grouping-output columns that are present in every grouping set;
  aggregate-result columns remain above the aggregate.
• Mark all parent filters unsupported for global aggregates and grouping sets
  containing an empty grouping set, while retaining aggregate-generated
  dynamic filters.
• Validate that every child returns one parent-filter result per input filter
  before positional remapping.
• Add physical optimizer and SQL regression coverage for mixed filter order,
  global-aggregate name collisions, constant predicates, and grouping sets.

Are these changes tested?

Yes:

• cargo test -p datafusion --test core_integration physical_optimizer::filter_pushdown
• cargo fmt --all -- --check
• cargo check -p datafusion-physical-plan -p datafusion-physical-optimizer
• git diff --check upstream/main
• Full GitHub CI, including Rust tests, clippy, and sqllogictests

Are there any user-facing changes?

There are no public API changes. Filter pushdown now preserves mixed aggregate
predicates correctly and avoids moving predicates across aggregates when doing
so could change empty-input results.

[discord9]

Hi, this PR fixes #22925 and is still waiting for review.

The only failing check is Circular Dependency Check, which appears to fail while compiling cargo-credential rather than from this PR's code. Could someone please rerun it or advise if I should do anything here?

Thanks!

[codecov-commenter]

Codecov Report

❌ Patch coverage is 90.47619% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 80.75%. Comparing base (ef94a85) to head (4e9f319).
⚠️ Report is 9 commits behind head on main.

Files with missing lines	Patch %	Lines
...tafusion/physical-optimizer/src/filter_pushdown.rs	75.00%	1 Missing ⚠️
datafusion/physical-plan/src/aggregates/mod.rs	94.11%	1 Missing ⚠️

Additional details and impacted files

@@ Coverage Diff @@## main #22926 +/- ##
==========================================
+ Coverage 80.70% 80.75% +0.04% 
==========================================
Files 1095 1096 +1 Lines 372554 373588 +1034 Branches 372554 373588 +1034 ==========================================
+ Hits 300665 301677 +1012 + Misses 53921 53906 -15 - Partials 17968 18005 +37 

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[neilconway]

Thank you for this PR! Overall, this is excellent work: clear diagnosis, minimal and correct fix, and good test coverage. I'm sorry it didn't get reviewed sooner -- a few minor suggestions but it should be good to land shortly.

[neilconway]

Looking good! Can you update the PR description, please? Currently refers to the old implementation; would also be good to elaborate on the pushdown change for global aggregates?

[discord9]

Updated the PR description to reflect the current from_child_with_allowed_indices implementation, the descriptor-length invariant, and why global aggregates / grouping sets containing () must retain parent filters above the aggregate. Thanks for the review.