feat: 新增外部数据源拉取与 DPO 输出格式支持

- 支持从 PostgreSQL 数据库拉取结构化数据作为训练来源
- 新增 DPO (Direct Preference Optimization) 输出类型
- 支持 chosen/rejected 字段的编辑、校验和发布
- 完善数据预处理切分逻辑和元数据管理
- 移除 OCR 扫描 PDF 功能，保持基础文本解析能力

Co-Authored-By: Claude <noreply@anthropic.com>

backend/tests/test_data_process_generation.py

@@ -88,6 +88,74 @@ def test_generate_model_records_uses_prompt_auth_and_stable_split() -> None:
     assert progress_updates == [(1, 1)]
 
 
+def test_generate_model_records_builds_native_dpo_pair() -> None:
+    def handler(request: httpx.Request) -> httpx.Response:
+        payload = json.loads(request.content)
+        system_prompt = payload["messages"][0]["content"]
+        assert '"chosen"' in system_prompt
+        assert '"rejected"' in system_prompt
+        assert "直接偏好优化" in system_prompt
+        return httpx.Response(
+            200,
+            json={
+                "choices": [{
+                    "message": {
+                        "content": json.dumps({
+                            "items": [{
+                                "instruction": "系统如何处理扫描 PDF？",
+                                "input": "",
+                                "chosen": "仅在没有文本层时调用 OCR，并保留页码。",
+                                "rejected": "所有 PDF 都重复执行 OCR。",
+                            }],
+                        }, ensure_ascii=False),
+                    },
+                }],
+            },
+        )
+
+    records = generate_model_records(
+        [{"id": "preview-dpo", "edited_content": "扫描 PDF 缺少文本层时执行 OCR。"}],
+        model={"name": "model", "api_url": "https://model.example/v1"},
+        config={"output_type": "dpo", "generation_retries": 0},
+        task_id="task-dpo",
+        split={"train": 100, "validation": 0, "test": 0},
+        qa_pairs_per_item=1,
+        client=httpx.Client(transport=httpx.MockTransport(handler)),
+    )
+
+    assert records[0]["status"] == "valid"
+    assert records[0]["chosen"] == "仅在没有文本层时调用 OCR,并保留页码。"
+    assert records[0]["rejected"] == "所有 PDF 都重复执行 OCR。"
+    assert records[0]["output"] == records[0]["chosen"]
+
+
+def test_generate_model_records_rejects_equal_dpo_pair() -> None:
+    def handler(_request: httpx.Request) -> httpx.Response:
+        return httpx.Response(
+            200,
+            json={"choices": [{"message": {"content": json.dumps({
+                "items": [{
+                    "instruction": "问题",
+                    "chosen": "相同回答",
+                    "rejected": "相同回答",
+                }],
+            }, ensure_ascii=False)}}]},
+        )
+
+    records = generate_model_records(
+        [{"id": "preview-dpo-invalid", "edited_content": "来源"}],
+        model={"name": "model", "api_url": "https://model.example/v1"},
+        config={"output_type": "dpo", "generation_retries": 0},
+        task_id="task-dpo-invalid",
+        split={"train": 100, "validation": 0, "test": 0},
+        qa_pairs_per_item=1,
+        client=httpx.Client(transport=httpx.MockTransport(handler)),
+    )
+
+    assert records[0]["status"] == "invalid"
+    assert "chosen equals rejected" in records[0]["error"]
+
+
 def test_minimax_m3_uses_split_reasoning_and_completion_token_budget() -> None:
     requests: list[dict[str, object]] = []