Update changelog for RAPIDS AI validation

Author: lmeyerov

CHANGELOG.md

@@ -10,12 +10,14 @@ This project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.htm
 ### 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.
+- **RAPIDS / AI compatibility**: Hardened GPU AI/text/UMAP/DBSCAN paths across official RAPIDS `25.02-cuda12.8` and `26.02-cuda13` by routing fragile cudf host conversions through shared compat helpers, avoiding crash-prone cudf `query()` filtering on GPU text paths, and using the safer cuML UMAP input/binding path on the known-bad `25.x` DGX arm64 stack.
 
 ### 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`.
+- **Tests / RAPIDS AI**: Amplified direct regression coverage around the new compat helpers and stabilized the DGX-only DBSCAN parity comparison with seeded UMAP kwargs; confirmed `test-text`, `test-features`, `test-umap-learn-core`, and `test-dbscan` pass on official RAPIDS `25.02-cuda12.8` and `26.02-cuda13`.
 
 ### Docs
 - **GFQL**: Show Cypher string syntax above the fold in "10 Minutes to GFQL" and "Overview" pages so the first code a reader sees is familiar Cypher, not the native chain API.