feat(server): 扩展文档智能识别服务，新增Azure Document Intelligence集成和测试用例

server/tests/test_document_intelligence.py

@@ -64,3 +64,55 @@ def test_document_intelligence_service_uses_vlm_result_when_preview_available(mo
     assert insight.document_type == "taxi_receipt"
     assert insight.classification_source == "llm_vision"
     assert calls[0] == ("vlm",)
+
+
+def test_document_intelligence_extracts_larger_decimal_amount_from_multiple_candidates() -> None:
+    insight = build_document_insight(
+        filename="taxi-amount.png",
+        summary="滴滴出行电子行程单",
+        text="滴滴出行 支付金额 1 元，实付 13.4 元，订单号 12345678",
+    )
+
+    assert any(field.label == "金额" and field.value == "13.4元" for field in insight.fields)
+
+
+def test_document_intelligence_service_uses_vlm_fields_to_correct_amount(monkeypatch) -> None:
+    def fake_complete(self, messages, *, slot_priority=("main", "backup"), max_tokens=500, temperature=0.2):
+        if slot_priority == ("vlm",):
+            return json.dumps(
+                {
+                    "document_type": "taxi_receipt",
+                    "scene_code": "transport",
+                    "scene_label": "交通票据",
+                    "expense_type": "transport",
+                    "confidence": 0.89,
+                    "evidence": ["图片主体为滴滴行程单，金额区域显示 13.4 元"],
+                    "fields": [
+                        {"key": "amount", "label": "金额", "value": "13.4"},
+                        {"key": "merchant_name", "label": "商户", "value": "滴滴出行"},
+                    ],
+                },
+                ensure_ascii=False,
+            )
+        return None
+
+    monkeypatch.setattr(RuntimeChatService, "complete", fake_complete)
+
+    engine = create_engine(
+        "sqlite+pysqlite:///:memory:",
+        connect_args={"check_same_thread": False},
+        poolclass=StaticPool,
+    )
+    session = sessionmaker(bind=engine, autoflush=False, autocommit=False)()
+    try:
+        insight = DocumentIntelligenceService(session).build_document_insight(
+            filename="didi-corrected.png",
+            summary="滴滴出行电子行程单",
+            text="滴滴出行 支付金额 1 元 订单号 12345678",
+            preview_data_url="data:image/png;base64,ZmFrZQ==",
+        )
+    finally:
+        session.close()
+
+    assert any(field.label == "金额" and field.value == "13.4元" for field in insight.fields)
+    assert any("大模型复核结果修正" in warning for warning in insight.warnings)