feat: IM式跨端实时同步——message_persisted 广播 + 前端按id去重追加

- history.Store 新增 AppendFull（返回含 id/createdAt 的完整消息），Append 签名不变
- session_orchestrator 加可空 OnPersisted；persistUserMessages/persistAssistantReply 落库后回调广播
- server.go newOrchestrator 接 OnPersisted → hub.broadcast message_persisted（thinking 剥除）
- 前端 ws.ts 加 message_persisted case（仅 activeSessionId 匹配才处理）；appendOrUpdateMessage 升级为 id 精确upsert+内容/时间归一
- 实测：双 WS，A stream 发消息，B 免刷新收到 user+assistant 各1条（count=2 PASS）
- 红线: 身份仍靠 actor 元数据，不靠提示词；用户/助手消息、本端/异端统一路径

internal/services/gateway/session_orchestrator.go

@@ -7,6 +7,7 @@ import (
 	"sync"
 	"time"
 
+	"git.yeij.top/AskaEth/Cyrene_v2/internal/runtime/history"
 	"git.yeij.top/AskaEth/Cyrene_v2/internal/runtime/loop"
 	"git.yeij.top/AskaEth/Cyrene_v2/internal/runtime/model"
 )
@@ -108,6 +109,11 @@ type OrchestratorOptions struct {
 	// When nil every Append is skipped, so the orchestrator runs without storage
 	// (tests / a gateway built without a history store).
 	History HistoryStore
+	// OnPersisted, when non-nil, is called with the complete stored message right
+	// after a user or assistant message is written, so the transport can broadcast
+	// a `message_persisted` event for cross-end realtime sync (design/16 §4.2).
+	// Dropped when nil (tests / a gateway built without a transport hook).
+	OnPersisted func(history.Message)
 }
 
 // sessionState is the per-session message state (design/15 §2). Every mutation
@@ -115,7 +121,7 @@ type OrchestratorOptions struct {
 // lock so a slow model call never blocks Submit.
 type sessionState struct {
 	mode    Mode
-	gen     uint64          // generation: bumped on interrupt and on a fresh top-level turn
+	gen     uint64 // generation: bumped on interrupt and on a fresh top-level turn
 	cancel  context.CancelFunc
 	pending []pendingMsg
 	tail    []model.Message // bounded conversation tail for context continuation
@@ -134,14 +140,15 @@ type pendingMsg struct {
 // actual run off to a goroutine. It is the single owner of "is this session busy"
 // for the chat path; the loop/task/subagent engines are reused, not re-created.
 type Orchestrator struct {
-	mu         sync.Mutex
-	states     map[string]*sessionState
-	run        TurnRunner
-	emit       StateSink
-	log        *slog.Logger
-	history    HistoryStore
-	maxPending int
-	maxTail    int
+	mu          sync.Mutex
+	states      map[string]*sessionState
+	run         TurnRunner
+	emit        StateSink
+	log         *slog.Logger
+	history     HistoryStore
+	onPersisted func(history.Message)
+	maxPending  int
+	maxTail     int
 }
 
 // NewOrchestrator builds an orchestrator from opts. Run is required: it is the
@@ -161,13 +168,14 @@ func NewOrchestrator(opts OrchestratorOptions) *Orchestrator {
 		opts.Log = slog.New(slog.NewTextHandler(io.Discard, nil))
 	}
 	return &Orchestrator{
-		states:     make(map[string]*sessionState),
-		run:        opts.Run,
-		emit:       opts.Emit,
-		log:        opts.Log,
-		history:    opts.History,
-		maxPending: maxPending,
-		maxTail:    maxTail,
+		states:      make(map[string]*sessionState),
+		run:         opts.Run,
+		emit:        opts.Emit,
+		log:         opts.Log,
+		history:     opts.History,
+		onPersisted: opts.OnPersisted,
+		maxPending:  maxPending,
+		maxTail:     maxTail,
 	}
 }
 
@@ -414,6 +422,14 @@ func (o *Orchestrator) emitState(sessionID, typ string, data map[string]any) {
 	}
 }
 
+// notifyPersisted hands a just-persisted message to the transport hook when one is
+// wired, so the gateway can broadcast `message_persisted` for cross-end sync.
+func (o *Orchestrator) notifyPersisted(msg history.Message) {
+	if o.onPersisted != nil {
+		o.onPersisted(msg)
+	}
+}
+
 // boundedTail returns a fresh copy of the last maxTail messages, so continuation
 // stays incremental and bounded (design/15 §2.2, §4) and never aliases the input
 // slice or another state's tail.