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/extractor_test.go

@@ -0,0 +1,79 @@
+package main
+
+import (
+	"path/filepath"
+	"testing"
+
+	"git.yeij.top/AskaEth/Cyrene_v2/internal/config"
+	"git.yeij.top/AskaEth/Cyrene_v2/internal/runtime/memory"
+	modelpkg "git.yeij.top/AskaEth/Cyrene_v2/internal/runtime/model"
+)
+
+// testRuntimeLLM builds a provider stack from env alone (the pointed-at
+// models.json does not exist), so a test controls the active group through the
+// flat LLM config exactly as a keyless .env boot would.
+func testRuntimeLLM(t *testing.T, env config.LLMConfig) *modelpkg.RuntimeLLM {
+	t.Helper()
+	return modelpkg.NewRuntimeLLM(filepath.Join(t.TempDir(), "models.json"), env, testLog())
+}
+
+// TestBuildMemoryExtractor_LLMWhenCredentials proves a configured provider group
+// selects the LLM extractor (design/03 §1).
+func TestBuildMemoryExtractor_LLMWhenCredentials(t *testing.T) {
+	t.Parallel()
+	a := &app{
+		log: testLog(),
+		llm: testRuntimeLLM(t, config.LLMConfig{
+			BaseURL: "https://api.example.com/v1",
+			APIKey:  "sk-test",
+			Model:   "gpt-4o-mini",
+		}),
+	}
+	if _, ok := a.buildMemoryExtractor("persona-a").(*memory.LLMExtractor); !ok {
+		t.Fatalf("extractor = %T, want *memory.LLMExtractor (credentials configured)", a.buildMemoryExtractor("persona-a"))
+	}
+}
+
+// TestBuildMemoryExtractor_ManualWhenNoCredentials proves the keyless fallback: a
+// boot without an API key (and without a local endpoint) keeps the deterministic
+// ManualExtractor instead of firing a doomed model call on every turn.
+func TestBuildMemoryExtractor_ManualWhenNoCredentials(t *testing.T) {
+	t.Parallel()
+	a := &app{log: testLog(), llm: testRuntimeLLM(t, config.LLMConfig{})}
+	if _, ok := a.buildMemoryExtractor("persona-a").(*memory.ManualExtractor); !ok {
+		t.Fatalf("extractor = %T, want *memory.ManualExtractor (no usable llm group)", a.buildMemoryExtractor("persona-a"))
+	}
+
+	bare := &app{log: testLog()}
+	if _, ok := bare.buildMemoryExtractor("persona-a").(*memory.ManualExtractor); !ok {
+		t.Fatalf("extractor = %T, want *memory.ManualExtractor (nil provider stack)", bare.buildMemoryExtractor("persona-a"))
+	}
+}
+
+// TestLLMGroupUsable_KeylessLocalEndpoint locks the ollama-style case: a local
+// OpenAI-compatible server needs no API key, so extraction stays LLM-backed.
+func TestLLMGroupUsable_KeylessLocalEndpoint(t *testing.T) {
+	t.Parallel()
+	cases := []struct {
+		name  string
+		env   config.LLMConfig
+		ready bool
+	}{
+		{name: "api key", env: config.LLMConfig{BaseURL: "https://api.example.com/v1", APIKey: "sk", Model: "m"}, ready: true},
+		{name: "loopback keyless", env: config.LLMConfig{BaseURL: "http://127.0.0.1:11434/v1", Model: "llama3"}, ready: true},
+		{name: "localhost keyless", env: config.LLMConfig{BaseURL: "http://localhost:1234/v1", Model: "m"}, ready: true},
+		{name: "private keyless", env: config.LLMConfig{BaseURL: "http://192.168.1.20:11434/v1", Model: "m"}, ready: true},
+		{name: "remote keyless", env: config.LLMConfig{BaseURL: "https://api.example.com/v1", Model: "m"}, ready: false},
+		{name: "no model", env: config.LLMConfig{BaseURL: "https://api.example.com/v1", APIKey: "sk"}, ready: false},
+		{name: "no base url", env: config.LLMConfig{APIKey: "sk", Model: "m"}, ready: false},
+	}
+	for _, tc := range cases {
+		tc := tc
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			if got := llmGroupUsable(testRuntimeLLM(t, tc.env)); got != tc.ready {
+				t.Fatalf("llmGroupUsable = %v, want %v", got, tc.ready)
+			}
+		})
+	}
+}