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)

cmd/cyrene/wire.go

@@ -6,7 +6,6 @@ import (
 	"encoding/json"
 	"fmt"
 	"log/slog"
-	"os"
 	"path/filepath"
 	"strings"
 	"time"
@@ -117,6 +116,11 @@ type app struct {
 	// satisfied by *gateway.Server (BroadcastMood); held as an interface so the
 	// composition root is unit-testable with a fake sink.
 	moodSink moodEmitter
+	// traceWriter fans finished trace spans out to the JSONL stderr recorder and,
+	// once the gateway exists, to the gateway's observability buffer behind
+	// GET /api/v1/admin/observability/trace (design/19 §2). It is held so
+	// wireFrontend can register the gateway sink on the already-running tracer.
+	traceWriter *traceFanout
 }
 
 // capacityEmitter is the gateway-facing sink for capacity_awaken broadcasts.
@@ -360,7 +364,10 @@ func (a *app) wireRuntime() error {
 	personaID := a.defaultPersonaID()
 	store := a.memStore
 
-	tracer := trace.New(trace.NewJSONRecorder(os.Stderr))
+	// Spans fan out to the JSONL stderr recorder and — once wireFrontend has built
+	// the gateway — to its observability buffer (cmd/cyrene/observability.go).
+	a.traceWriter = newAppTraceWriter()
+	tracer := trace.New(a.traceWriter)
 	// Realtime flow visualization (design/19 §1): the publisher accumulates the
 	// loop's trace spans into a per-session flow graph and publishes
 	// ws.graph_update on each span transition. It is gated to the owner's web
@@ -413,7 +420,7 @@ func (a *app) wireRuntime() error {
 		OnTurnEnd:     a.onTurnEnd,
 		Memory:        store,
 		Retriever:     a.retriever,
-		Extractor:     memory.NewManualExtractor(personaID, "owner"),
+		Extractor:     a.buildMemoryExtractor(personaID),
 		Checkpoint:    checkpoint,
 		Tracer:        tracer,
 		Flow:          flowPub,
@@ -738,6 +745,10 @@ func (a *app) wireFrontend() error {
 	}
 	gw.SetVoice(a.voice)
 	a.gw = gw // assign BEFORE wiring hooks below, which capture a.gw / gw (nil-receiver panic otherwise)
+	// Observability trace sink (design/19 §2): now that the gateway exists, its
+	// ring buffer must be fed by the loop's spans, or the admin trace/metrics
+	// endpoints stay empty.
+	wireGatewayTraceSink(a, gw)
 	if a.channelBridge != nil {
 		// Route the channel bridge's owner-private persisted messages into the same
 		// `message_persisted` broadcast the orchestrator uses, so the Web client mirrors