[python] Reuse gathered window values across aggregations

[QuakeWang]

Purpose

Window joins already deduplicate row IDs before fetching payloads, but each aggregation rebuilds Arrow indices and gathers the same source column again.

Gather each source column once per window and reuse it across aggregations. Release gathered values after processing that window and column, preserving output order, empty/null behavior, output types, and mean precision and overflow handling.

Tests

• Regression test verifies reduced take calls, interleaved outputs, overlapping windows, and empty/null behavior.
• All 67 multimodal temporal tests passed.
• Flake8, license headers, and git diff --check passed.

[JingsongLi]

Requirement fit: SUPPORTED. Implementation: CLEAN.

Reviewed 3de2fe3742d4. A window with several aggregations over one source column currently repeats the same Arrow gather. Sharing that gathered selection has a direct path through the public window join and a small implementation scope. Output ordering, null/empty windows and mean precision behavior remain intact.

Validation: all 67 multimodal temporal tests and 48 subtests passed. Current head CI is green. This confirms behavior and gather reuse; it does not quantify end-to-end latency savings.

No actionable implementation regression found in this review.