feat: 知识批量沉淀改为后台任务,避免 nginx 504

- /ingest/threads/batch 新增 background 参数(默认 true),立即返回 {queued:true}
- 抽取循环抽出 _run_batch_ingest,后台路径显式传入 created_by
- 前端知识库页与输入框、线程 hooks/types 配套调整
- config.yaml 上下文压缩测试值

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

offline-backend-20260512/backend/packages/harness/deerflow/agents/lead_agent/agent.py

@@ -54,12 +54,19 @@ def _resolve_model_name(requested_model_name: str | None = None, *, app_config:
     return default_model_name
 
 
-def _create_summarization_middleware(*, app_config: AppConfig | None = None) -> DeerFlowSummarizationMiddleware | None:
-    """Create and configure the summarization middleware from config."""
+def _create_summarization_middleware(*, app_config: AppConfig | None = None, force_enabled: bool = False) -> DeerFlowSummarizationMiddleware | None:
+    """Create and configure the summarization middleware from config.
+
+    Args:
+        force_enabled: When True, build the middleware regardless of the config
+            ``summarization.enabled`` flag. The per-request 上下文压缩 toggle
+            (chat input box, default off) drives this — the config file no longer
+            owns the on/off switch, only the tuning (trigger/keep/...).
+    """
     resolved_app_config = app_config or get_app_config()
     config = resolved_app_config.summarization
 
-    if not config.enabled:
+    if not (config.enabled or force_enabled):
         return None
 
     # Prepare trigger parameter
@@ -268,8 +275,13 @@ def _build_middlewares(
     if cfg.get("rag_mode_enabled", False):
         middlewares.append(RagCitationMiddleware())
 
-    # Add summarization middleware if enabled
-    summarization_middleware = _create_summarization_middleware(app_config=resolved_app_config)
+    # Add summarization middleware. The on/off switch now lives in the chat UI
+    # (上下文压缩 toggle, default off) and arrives as ``summarization_enabled`` in
+    # the runtime config; the config file only supplies tuning. force_enabled
+    # lets the per-request toggle turn it on even when config.enabled is false,
+    # while non-UI paths (channels/scheduled) still fall back to config.enabled.
+    summarization_requested = cfg.get("summarization_enabled", False)
+    summarization_middleware = _create_summarization_middleware(app_config=resolved_app_config, force_enabled=summarization_requested)
     if summarization_middleware is not None:
         middlewares.append(summarization_middleware)