feat: 重构报销单AI预审流程并添加平台风险规则引擎

- 将AI验审改为AI预审，高风险不再拦截而是随单流转给审批人复核
- 新增平台风险规则评估引擎，支持事由过短、票据异常、重复发票等多种评估器
- 用户上下文增加部门信息(department_name)，认证流程同步关联组织架构
- 规则scenario_json改为中文标签(差旅/费用科目)，统一场景分类
- 新增orchestrator审核流程测试用例
- 前端更新审计视图、差旅报销等相关页面

server/src/app/services/orchestrator.py

@@ -173,9 +173,11 @@ class OrchestratorService:
                 task_asset=task_asset,
             )
             selected_capability_codes = self._flatten_capability_codes(capabilities)
-            requires_confirmation = (
-                ontology.permission.level == AgentPermissionLevel.APPROVAL_REQUIRED.value
-            )
+            is_expense_review_action = self._is_expense_review_action(context_json)
+            requires_confirmation = (
+                ontology.permission.level == AgentPermissionLevel.APPROVAL_REQUIRED.value
+                and not is_expense_review_action
+            )
 
             route_json = {
                 "orchestrated_by": AgentName.ORCHESTRATOR.value,
@@ -526,7 +528,11 @@ class OrchestratorService:
                 failed_tool_count=1 if degraded else 0,
             )
 
-        next_step = self._resolve_next_step(ontology, payload.source)
+        next_step = self._resolve_next_step(
+            ontology,
+            payload.source,
+            context_json=context_json,
+        )
         if next_step == "query_database":
             tool_payload, degraded = self._invoke_tool(
                 run_id=run_id,
@@ -662,9 +668,9 @@ class OrchestratorService:
             "degraded": True,
         }
 
-        if ontology.scenario == "expense":
-            tool_type = AgentToolType.DATABASE.value
-            tool_name = "database.expense_claims.save_or_submit"
+        if ontology.scenario == "expense" or self._is_expense_review_action(context_json):
+            tool_type = AgentToolType.DATABASE.value
+            tool_name = "database.expense_claims.save_or_submit"
             executor = lambda: self.expense_claim_service.save_or_submit_from_ontology(
                 run_id=run_id,
                 user_id=payload.user_id,
@@ -781,10 +787,17 @@ class OrchestratorService:
             failed_tool_count=failed_tool_count,
         )
 
-    @staticmethod
-    def _resolve_next_step(ontology: OntologyParseResult, source: str) -> str:
-        if ontology.clarification_required:
-            return "ask_clarification"
+    @staticmethod
+    def _resolve_next_step(
+        ontology: OntologyParseResult,
+        source: str,
+        *,
+        context_json: dict[str, Any] | None = None,
+    ) -> str:
+        if OrchestratorService._is_expense_review_action(context_json or {}):
+            return "create_draft"
+        if ontology.clarification_required:
+            return "ask_clarification"
         if ontology.intent == "draft":
             return "create_draft"
         if ontology.scenario == "knowledge" or ontology.intent == "explain":
@@ -793,7 +806,18 @@ class OrchestratorService:
             return "run_rule"
         if ontology.intent in {"query", "compare"}:
             return "query_database"
-        return "create_draft"
+        return "create_draft"
+
+    @staticmethod
+    def _is_expense_review_action(context_json: dict[str, Any]) -> bool:
+        review_action = str((context_json or {}).get("review_action") or "").strip()
+        return review_action in {
+            "save_draft",
+            "next_step",
+            "edit_review",
+            "link_to_existing_draft",
+            "create_new_claim_from_documents",
+        }
 
     @staticmethod
     def _flatten_capability_codes(