Fix an apparent mistake in GraphBuilder.add_child (#879)

* Fix an apparent mistake in GraphBuilder.add_child

While working on C++ stand-alone code executing what `test_graph.py`
does in gh-843, I noticed that `add_child` passes dependendencies
extracted from capturing stream inconsistently with num_dependencies
parameter obtained in the same cuStreamGetCaptureInfo call.

Incidentally, after correcting this error, I can no longer reproduce
errors reported in gh-843

* Implemented fix to work with both CTK 12.9 and CTK 13.0

Author: oleksandr-pavlyk

cuda_core/cuda/core/experimental/_graph.py

@@ -683,11 +683,14 @@ def add_child(self, child_graph: GraphBuilder):
             driver.cuStreamGetCaptureInfo(stream_handle)
         )
 
+        # See https://github.com/NVIDIA/cuda-python/pull/879#issuecomment-3211054159
+        # for rationale
+        deps_info_trimmed = deps_info_out[:num_dependencies_out]
         deps_info_update = [
             [
                 handle_return(
                     driver.cuGraphAddChildGraphNode(
-                        graph_out, deps_info_out[0], num_dependencies_out, child_graph._mnff.graph
+                        graph_out, *deps_info_trimmed, num_dependencies_out, child_graph._mnff.graph
                     )
                 )
             ]