Document GFQL GPU optimization work

Author: lmeyerov

CHANGELOG.md

@@ -7,6 +7,15 @@ This project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.htm
 
 ## [Development]
 <!-- Do Not Erase This Section - Used for tracking unreleased changes -->
+### Fixed
+- **GFQL / GPU traversal**: Added a narrow one-hop undirected `hop()` fast path that avoids doubled edge-pair materialization for the common no-predicate traversal shape, reducing warm GPU GFQL search cost on benchmarked workloads.
+- **cuGraph projection**: `to_cugraph()` now auto-skips cuGraph renumbering only for already-safe zero-based contiguous signed integer IDs, with targeted regression coverage for sparse, nullable, string, offset, and disconnected topologies.
+
+### Added
+- **Benchmarks / RAPIDS GPU**: Added official-image GPU benchmark runners for GFQL stage breakdowns and direct `to_cugraph()` measurement across RAPIDS/CUDA variants, plus synthetic graph families that isolate sparse integer vs repeated string-ID graph-build behavior.
+
+### Changed
+- **GFQL / RAPIDS perf triage**: Added a pure cuDF/cuGraph reproducer for the RAPIDS `25.02 -> 26.02` string-ID graph-build regression and tracked follow-ups in issues `#977` and `#978`.
 
 ## [0.53.7 - 2026-03-29]