fix(translate): исправление INSERT + выгрузка CSV для неуспешных запусков

**Исправления ошибок:**
- superset_executor: убран несуществующий fallback endpoint /api/v1/sql_lab/execute/
  (правильный URL /api/v1/sqllab/execute/ подтверждён браузерным запросом Superset)
  Ошибка 500 (permission denied) больше не маскируется 404
- orchestrator_run_completion: run.status=FAILED при провале insert
  (раньше всегда ставился COMPLETED). timeout тоже treated as failure
- test_orchestrator: assertion обновлён COMPLETED→FAILED

**Новая функция — выгрузка CSV:**
- GET /api/translate/runs/{run_id}/failed.csv — возвращает CSV с успешно
  переведёнными, но не вставленными записями (status=SUCCESS)
- Колонки: record_id, source_*, target_sql, source_data (JSON), языки, error_message
- Метаданные запуска (run_status, insert_status, run_error) в конце файла
- Кнопки скачивания в 3 местах фронта:
  - История запусков (при status=FAILED)
  - TranslationRunProgress (insert_failed)
  - TranslationRunResult (failed/partial/insert_failed)
- downloadFailedCsv() через fetchApiBlob + Blob URL

backend/src/plugins/translate/__tests__/test_orchestrator.py

@@ -420,7 +420,7 @@ class TestTranslationOrchestrator:
     # endregion test_execute_run_no_job
 
     # region test_execute_run_insert_failure [TYPE Function]
-    # @BRIEF SQL generation/insert returns failure, run still completes but with error logged.
+    # @BRIEF SQL generation/insert returns failure — run marked FAILED with error logged.
     def test_execute_run_insert_failure(self, mock_job: MagicMock) -> None:
         db = MagicMock()
         config_manager = MagicMock()
@@ -464,7 +464,7 @@ class TestTranslationOrchestrator:
                  patch.object(engine, 'event_log'):
                 result = orch.execute_run(run)
 
-        assert result.status == "COMPLETED"
+        assert result.status == "FAILED"
         assert result.insert_status == "failed"
         assert "timeout" in (result.error_message or "")
 

backend/src/plugins/translate/orchestrator_run_completion.py

@@ -106,9 +106,17 @@ def complete_success(
     run.insert_status = insert_result.get("status")
     run.superset_execution_id = str(insert_result.get("query_id") or "")
     run.superset_execution_log = insert_result
-    if run.status != "FAILED":
+    # Treat "failed" and "timeout" as terminal failures.
+    # "skipped" (no records) and "success" are non-failure outcomes.
+    if insert_result.get("status") in ("failed", "timeout"):
+        run.status = "FAILED"
+        if insert_result.get("error_message"):
+            run.error_message = insert_result["error_message"]
+        elif not run.error_message:
+            run.error_message = f"SQL insert phase {insert_result.get('status')}"
+    elif run.status != "FAILED":
         run.status = "COMPLETED"
-    if insert_result.get("error_message"):
+    if insert_result.get("error_message") and not run.error_message:
         run.error_message = insert_result["error_message"]
     run.completed_at = datetime.now(UTC)
     db.flush()

backend/src/plugins/translate/superset_executor.py

@@ -173,36 +173,30 @@ class SupersetSqlLabExecutor:
                 "client_id": f"trl-{uuid.uuid4().hex[:7]}",  # max 11 chars for Superset varchar(11)
             }
 
-            # Try multiple SQL Lab endpoints — some Superset versions use /sqllab/execute/,
-            # others use /sql_lab/execute/ (with underscore)
-            candidate_endpoints = ["/api/v1/sqllab/execute/", "/api/v1/sql_lab/execute/"]
-            response = None
-            last_error = None
-            for endpoint in candidate_endpoints:
-                try:
-                    response = client.network.request(
-                        method="POST",
-                        endpoint=endpoint,
-                        data=json.dumps(payload),
-                        headers={"Content-Type": "application/json"},
-                    )
-                    if isinstance(response, dict) and response:
-                        logger.reason("SQL Lab endpoint succeeded", extra={
-                            "endpoint": endpoint,
-                        })
-                        break
-                except Exception as ep_err:
-                    last_error = ep_err
-                    logger.explore("SQL Lab endpoint failed, trying next", extra={
-                        "error": str(ep_err),
-                        "endpoint": endpoint,
-                    })
-                    response = None
-
-            if response is None:
-                error_msg = f"All SQL Lab endpoints failed. Last error: {last_error}"
-                logger.explore("SQL Lab execute failed", extra={"error": error_msg})
-                raise ValueError(f"Superset SQL Lab execute failed: {error_msg}")
+            # Use the canonical Superset SQL Lab execute endpoint.
+            # NOTE: Some older docs reference /sql_lab/execute/ (with underscore),
+            # but this endpoint does NOT exist in current Superset versions.
+            # The browser UI uses /api/v1/sqllab/execute/ — confirmed by
+            # fetch() calls in Superset SQL Lab.
+            endpoint = "/api/v1/sqllab/execute/"
+            try:
+                response = client.network.request(
+                    method="POST",
+                    endpoint=endpoint,
+                    data=json.dumps(payload),
+                    headers={"Content-Type": "application/json"},
+                )
+                logger.reason("SQL Lab endpoint succeeded", extra={
+                    "endpoint": endpoint,
+                })
+            except Exception as ep_err:
+                logger.explore("SQL Lab execute failed", extra={
+                    "error": str(ep_err),
+                    "endpoint": endpoint,
+                })
+                raise ValueError(
+                    f"Superset SQL Lab execute failed: {ep_err}"
+                ) from ep_err
 
             # Parse response — try multiple known Superset response formats
             result = response if isinstance(response, dict) else {}