feat(investigation): preserve cumulative reference graphs

Accumulate investigation paths and preserve failed attempts across rounds. Keep graph annotations separate from execution state, synchronize the API contract, and consolidate documentation.

tests/unit/test_investigation_diagram_status.py

@@ -28,7 +28,7 @@ def test_waiting_and_unknown_node_states_survive_progress_and_final_report():
         stage="final", outcome="concluded", data_gaps=[], diagram=diagram,
     )
     assert report.diagram == progress.diagram
-    assert [node.status for node in report.diagram.node_refs] == ["completed", "waiting", None, None]
+    assert [node.status for node in report.diagram.node_refs] == ["recorded", "needs_input", None, None]
     assert "status" not in report.diagram.node_refs[-1].model_fields_set
 
 
@@ -43,7 +43,7 @@ def test_class_attached_nodes_are_valid_declarations_without_accepting_partial_i
             "node_refs": [{"node_id": "QUERY", "label": "核查", "status": "running"},
                           {"node_id": "REVIEW", "label": "复核", "status": "pending"}]},
     }}
-    assert AgentArtifactsCommand.model_validate(payload).progress.diagram.node_refs[0].status == "running"
+    assert AgentArtifactsCommand.model_validate(payload).progress.diagram.node_refs[0].status == "reference"
     payload["progress"]["diagram"]["node_refs"][0]["node_id"] = "QUER"
     with pytest.raises(ValidationError, match="QUER"):
         AgentArtifactsCommand.model_validate(payload)