[FEA] cuda.core: expose graph and graph-node identity/introspection APIs (CUDA 13.1)

[leofang]

Summary

CUDA 13.1 adds stable identifiers and navigation for graphs and nodes; CUDA 13.3 adds a related
function attribute. None are exposed by cuda.core.graph today.

Motivation: correlating cuda.core graph objects with profiler/tooling output (toolsNodeId),
diagnostics and logging, keying per-node bookkeeping for the update work in #2352 / #2354, and
navigating from a node back to its owning graph.

Underlying C APIs to cover

C API	Purpose
cuGraphGetId(CUgraph, unsigned int* graphId)	stable ID of a graph definition
cuGraphExecGetId(CUgraphExec, unsigned int* graphId)	stable ID of an executable graph
cuGraphNodeGetLocalId(CUgraphNode, unsigned int* nodeId)	node ID local to its graph
cuGraphNodeGetToolsId(CUgraphNode, unsigned long long* toolsNodeId)	node ID as reported by profiling/tooling interfaces
cuGraphNodeGetContainingGraph(CUgraphNode, CUgraph*)	navigate from node to owning graph
CU_FUNC_ATTRIBUTE_DEVICE_NODE_UPDATE_SUPPORTED (13.3)	per-function attribute: kernel may perform device-side node updates

Design sketch (draft — needs design-meeting review)

Important

Starting point only, not a settled design — review in the cuda.core design meeting.

Read-only properties, no new objects:

• .id on the graph-definition and executable-graph wrappers (mapping to
  cuGraphGetId / cuGraphExecGetId respectively);
• .local_id, .tools_id, .containing_graph on the node wrappers in cuda.core.graph;
• the 13.3 function attribute joins the existing kernel-attribute surface.

Open questions for the meeting:

1. Exact mapping onto cuda.core's graph classes (which wrapper owns CUgraph vs CUgraphExec).
2. Plain int IDs vs. small typed wrappers.
3. containing_graph: return an existing wrapper (identity-cached) or a fresh non-owning one —
   consistent with from_handle constructors need to allow managing ownership #1187 / Inconsistent lifetime management across from_handle / from_* APIs #1989 lifetime conventions.

References

• Driver docs (Graph Management): https://docs.nvidia.com/cuda/cuda-driver-api/group__CUDA__GRAPH.html
• Found during the CUDA 12.8 → 13.3 bindings vs. cuda.core gap sweep (2026-07-14)

-- Leo's bot