fix(graph): treat cancel_node as control flow, not fatal error

[Zelys-DFKH]

Description

When cancel_node is set in a BeforeNodeCallEvent hook, the intent is to skip a node and keep the graph running. What actually happens: _execute_node raises RuntimeError after emitting MultiAgentNodeCancelEvent, which propagates through _execute_nodes_parallel and kills the graph. Downstream nodes never run, the overall status becomes FAILED, and on interrupt-and-resume workflows the session is left with no recovery path — the next call starts a fresh graph instead of continuing.

The fix marks the skipped node as Status.COMPLETED, records it in state.completed_nodes, state.results, and state.execution_order, yields MultiAgentNodeStopEvent, and returns. _find_newly_ready_nodes picks up downstream nodes from completed_batch as usual. get_agent_results() returns [] for the RuntimeError result, so downstream nodes see no output from the skipped node. No new status enum needed. Also corrected the cancel_node docstring, which still said "status FAILED".

Related Issues

Resolves #2240

Documentation PR

N/A

Type of Change

Bug fix

Testing

Ran hatch run prepare. Updated test_graph_cancel_node (parametrized ×2) to check for normal completion instead of raising. Added test_graph_cancel_node_downstream_executes: cancels step_a via hook and verifies step_b still runs, with both nodes in completed_nodes and results. No public API changes, so no impact on consuming repos.

• I ran hatch run prepare

Checklist

• I have read the CONTRIBUTING document
• I have added any necessary tests that prove my fix is effective or my feature works
• I have updated the documentation accordingly
• I have added an appropriate example to the documentation to outline the feature, or no new docs are needed
• My changes generate no new warnings
• Any dependent changes have been merged and published

---

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.