refactor(server): user_agent/steward/ocr 等服务重构并适配关联任务

- user_agent 拆分 application/locations/knowledge/response/review 四个子模块，接入申请位置语义与关联草稿分支
- steward planner/runtime/slot/plan_builder 决策链路重构，travel_reimbursement_calculator/orchestrator_expense_query 适配
- ocr/document_preview/document_intelligence/receipt_folder 复用预览与资产缓存，expense_claim_draft_flow/application_handoff 适配
- pyproject.toml 新增依赖，paddleocr bootstrap 脚本与 server_start.sh 调整
- 更新差旅/交通/通信等财务规则表，同步 document_intelligence/ocr/receipt_folder/user_agent 等测试

server/src/app/services/steward_slot_decision_agent.py

@@ -275,7 +275,7 @@ class StewardSlotDecisionAgent:
                 missing_fields=missing_fields,
                 question=self._build_fallback_question(field),
                 options=self._sanitize_options([], [field]),
-                rationale="模型字段决策暂不可用，我先按上游意图识别给出的本体缺口向你确认。",
+                rationale="模型字段决策暂不可用，我先按上游意图识别给出的本体缺口向您确认。",
                 model_call_traces=traces,
             )
         return StewardSlotDecisionResponse(
@@ -285,7 +285,7 @@ class StewardSlotDecisionAgent:
             missing_fields=[],
             question="",
             options=[],
-            rationale="当前任务没有上游标记的关键字段缺口，可以先生成核对结果供你确认。",
+            rationale="当前任务没有上游标记的关键字段缺口，可以先生成核对结果供您确认。",
             model_call_traces=traces,
         )
 
@@ -293,7 +293,7 @@ class StewardSlotDecisionAgent:
     def _build_fallback_question(field: str) -> str:
         label = FIELD_CATALOG.get(field, {}).get("label") or field
         if field == "transport_mode":
-            return "请问你这次打算怎么出行？可以选择火车、飞机或轮船。"
+            return "请问您这次打算怎么出行？可以选择火车、飞机或轮船。"
         return f"当前还缺少{label}，请先补充后我再继续处理。"
 
     @staticmethod