GFQL/Cypher: audit and broaden cuDF coverage for temporal comparisons #1446

Summary

Follow-up from #1374 / #1442. The Temporal7 pandas regression table is broad, while the cuDF path added in #1442 is a focused smoke using a cuDF-backed graph. Audit the GFQL/Cypher temporal comparison coverage and broaden cuDF coverage where it is likely to catch engine-specific regressions.

Scope

• Reuse or factor the Temporal7 comparison case data so selected cases can run against both pandas and cuDF-backed graphs.
• Cover a bounded representative set across date, time, and datetime comparisons.
• Include literal-vs-constructor equality cases where cuDF execution may diverge from pandas.
• Audit nearby GFQL/Cypher regression tests for pandas-only coverage of engine-sensitive behavior.
• Keep runtime bounded; prefer a curated cuDF subset over blindly duplicating every pandas parametrized case.

Acceptance criteria

• New or refactored tests prove the selected cases run on actual cuDF-backed DataFrames, not merely inside a RAPIDS image.
• DGX validation is recorded for RAPIDS 25.02 and 26.02.
• Existing pandas temporal comparison coverage remains intact.
• The changelog records the coverage/audit improvement if code changes land.

Notes

#1442 already validated one cuDF-backed temporal comparison on DGX RAPIDS 25.02 and 26.02. This issue is for broader coverage symmetry and audit follow-through, not a blocker for #1442.