feat(runtime+admin): A2 LLM异步提取接线 + A4 Tracing接线 + A6通知后端聚合

A2 异步记忆提取:
- loop/memory.go: asyncExtractor接口(ExtractAsync)+extractMemoryAsync(60s超时,WithOutCancel,dispatch即返回,后台落库,失败仅Warn)+persistMemories共用
- run.go 三个回合落定点改异步提取; 删除run.go逐步提取(design/03'每轮后', 避免多轮工具N次LLM调用/重复写)
- cmd/cyrene/extractor.go buildMemoryExtractor: llmGroupUsable为真→LLMExtractor, 否则回落ManualExtractor(日志说明); keylessEndpoint放行loopback/私网

A4 可观测性 Tracing:
- cmd/cyrene/observability.go: newAppTraceWriter+traceFanout(多sink)+gatewaySpanSink(走gw.ObserveSpan,喂环形缓冲+LLM/subagent指标)
- wire.go 注入; 未改trace/gateway数据结构

A6 通知后端:
- internal/services/gateway/notifications.go(311行): GET /admin/notifications?type=&before=&limit= (adminAuth)
- 聚合已有源(obs.audit SQLite + s.backthink), createdAt倒序, id=source:id; 类型派生(channel/system/audit/backthink); 无新建表/无新Deps/wire零改动
- 诚实: backthink仅shouldPush计, 未注入源为空, 报错500, 未读游标后置
- 契约与前端Y一致: {id,type,title,body,source,createdAt}, before含游标(<=), type未知400
- Docs/api-client v0.33

全仓 build/vet/test绿(56包)+win/darwin/linux-arm64交叉编译绿; 新增测试(memory_async/extractor/observability/notifications)

internal/runtime/loop/run.go

@@ -18,8 +18,10 @@ import (
 // timeout and fed back as RoleTool messages so the model can continue.
 // A checkpoint is persisted at every step, every step is traced as a
 // state_transition span (with an llm_call child and tool_call spans for each
-// invocation), and after each step the memory Extractor runs with errors logged
-// rather than fatal.
+// invocation), and when the turn settles the memory Extractor runs over the
+// turn's conversation (design/03 §1: the LLM extractor on a background
+// goroutine, so extraction never delays the reply) with errors logged rather
+// than fatal.
 //
 // When a turn settles with a Result, the OnTurnEnd hook is called once for that
 // turn (never per step) and its mood label is recorded on Result.Mood, so the
@@ -84,7 +86,7 @@ func (l *Loop) Run(ctx context.Context, sessionID string, messages []model.Messa
 			res.Text = outcome.text
 			res.StopReason = outcome.stopReason
 			l.saveCheckpoint(ctx, sessionID, step, msgs, outcome.stopReason)
-			l.extractMemory(ctx, msgs, step)
+			l.extractMemoryAsync(ctx, msgs, step)
 			l.settleTurn(ctx, sessionID, msgs, res)
 			return res, nil
 		}
@@ -96,7 +98,6 @@ func (l *Loop) Run(ctx context.Context, sessionID string, messages []model.Messa
 		msgs = batch.messages
 		res.ToolCallsExecuted += batch.executed
 		l.saveCheckpoint(ctx, sessionID, step, msgs, agent.StopReasonToolCalls)
-		l.extractMemory(ctx, msgs, step)
 
 		// The critical-tool-loop safety net (or a terminating hook) stops the
 		// round without waiting for a real final answer.
@@ -104,7 +105,7 @@ func (l *Loop) Run(ctx context.Context, sessionID string, messages []model.Messa
 			res.Text = criticalToolLoopTerminalReason
 			res.StopReason = StopReasonToolLoopTerminated
 			l.saveCheckpoint(ctx, sessionID, step, msgs, StopReasonToolLoopTerminated)
-			l.extractMemory(ctx, msgs, step)
+			l.extractMemoryAsync(ctx, msgs, step)
 			l.settleTurn(ctx, sessionID, msgs, res)
 			return res, nil
 		}
@@ -114,7 +115,7 @@ func (l *Loop) Run(ctx context.Context, sessionID string, messages []model.Messa
 	res.Steps = l.maxRounds
 	res.StopReason = StopReasonMaxRounds
 	l.saveCheckpoint(ctx, sessionID, l.maxRounds, msgs, StopReasonMaxRounds)
-	l.extractMemory(ctx, msgs, l.maxRounds)
+	l.extractMemoryAsync(ctx, msgs, l.maxRounds)
 	l.settleTurn(ctx, sessionID, msgs, res)
 	return res, nil
 }