fix(graph): prevent reset_executor_state from corrupting MultiAgentBase state

[giulio-leone]

Problem

GraphNode.reset_executor_state() corrupts the executor state when the executor is a MultiAgentBase (e.g. a nested Graph).

The method checks hasattr(self.executor, "state") without distinguishing executor type, then unconditionally assigns AgentState(...). For MultiAgentBase executors whose .state is a GraphState, this overwrites it with the wrong type — breaking the nested executor.

Two call sites are affected:

• _execute_node: guarded by reset_on_revisit (triggers when revisiting a nested graph node in a cycle)
• deserialize_state: unguarded (triggers when restoring a completed run with no next_nodes_to_execute)

Root Cause

__post_init__ already has the correct guard:

if hasattr(self.executor, "state") and hasattr(self.executor.state, "get"):
    self._initial_state = AgentState(self.executor.state.get())

But reset_executor_state() was missing the hasattr(self.executor.state, "get") check:

if hasattr(self.executor, "state"):
    self.executor.state = AgentState(self._initial_state.get())  # Wrong for MultiAgentBase

GraphState doesn't have a .get() method, so the guard correctly skips it.

Fix

Add the same hasattr(self.executor.state, "get") guard to reset_executor_state(), matching the existing pattern in __post_init__.

Test

Added test_reset_executor_state_preserves_multiagent_graph_state — verifies that a GraphNode wrapping a MultiAgentBase with GraphState preserves the state type after reset_executor_state().

Fixes #1775

[giulio-leone]

Closing — focusing on higher-impact contributions.