feat(ai-core): trace instrumentation, LLM thinker, decay scheduler

- orchestrator: Tracer interface with spans for intent analysis, persona,
  subsession dispatch, context build, LLM streaming and each tool call;
  traces finish when the stream completes — the admin Trace waterfall
  finally receives data
- background: Thinker generates real LLM thoughts (post-chat + silence
  triggers, recent history as context, empty/overlong outputs discarded);
  thoughts broadcast to web clients as 'thinking' WS messages; pipeline
  postChatHook triggers post-chat thinking after each round
- memory: ApplyDecayAll + ListUserIDs with a 6h decay/purge cron in main
  (was implemented but never scheduled)

backend/services/ai-core/background/thinker.go

@@ -4,11 +4,23 @@ package background
 
 import (
 	"context"
+	"fmt"
 	"git.yeij.top/AskaEth/Cyrene_v2/internal/logger"
+	"strings"
 	"sync"
 	"time"
+
+	"git.yeij.top/AskaEth/Cyrene_v2/services/ai-core/model"
 )
 
+// LLMClient is the minimal LLM interface needed for thought generation.
+type LLMClient interface {
+	Chat(ctx context.Context, messages []model.LLMMessage) (*model.LLMResponse, error)
+}
+
+// HistoryProvider returns recent conversation messages for a session.
+type HistoryProvider func(sessionID string, limit int) []model.Message
+
 // Thinker manages autonomous background thinking.
 type Thinker struct {
 	mu              sync.Mutex
@@ -20,6 +32,9 @@ type Thinker struct {
 	thoughts        []Thought
 	stopCh          chan struct{}
 	postChatCh      chan string
+	llm             LLMClient
+	history         HistoryProvider
+	out             chan Thought
 }
 
 // Thought represents an autonomous thought generated by the AI.
@@ -39,9 +54,16 @@ func New(enabled bool, thinkInterval, silenceTimeout time.Duration) *Thinker {
 		thoughts:       make([]Thought, 0),
 		stopCh:         make(chan struct{}),
 		postChatCh:     make(chan string, 10),
+		out:            make(chan Thought, 10),
 	}
 }
 
+// SetLLM injects the LLM used to generate thoughts.
+func (t *Thinker) SetLLM(c LLMClient) { t.llm = c }
+
+// SetHistoryProvider injects a way to fetch recent conversation context.
+func (t *Thinker) SetHistoryProvider(fn HistoryProvider) { t.history = fn }
+
 // Start begins the background thinking loops.
 func (t *Thinker) Start(ctx context.Context) {
 	if !t.enabled {
@@ -80,12 +102,9 @@ func (t *Thinker) UpdateActivity() {
 	t.mu.Unlock()
 }
 
-// ThoughtChan returns a channel that emits new thoughts.
-// Not yet wired — will connect to NATS for event broadcasting.
-func (t *Thinker) ThoughtChan() chan Thought {
-	ch := make(chan Thought, 10)
-	// In future: push thoughts to this channel
-	return ch
+// ThoughtChan returns the channel that emits new thoughts.
+func (t *Thinker) ThoughtChan() <-chan Thought {
+	return t.out
 }
 
 func (t *Thinker) silenceLoop(ctx context.Context) {
@@ -126,9 +145,13 @@ func (t *Thinker) postChatLoop(ctx context.Context) {
 }
 
 func (t *Thinker) generateThought(trigger string, sessionID string) {
-	// Placeholder: in production this calls the LLM
+	content := t.think(trigger, sessionID)
+	if content == "" {
+		return
+	}
+
 	thought := Thought{
-		Content:   "[思考] 当前无主动想法",
+		Content:   content,
 		SessionID: sessionID,
 		CreatedAt: time.Now(),
 	}
@@ -140,9 +163,60 @@ func (t *Thinker) generateThought(trigger string, sessionID string) {
 	}
 	t.mu.Unlock()
 
+	select {
+	case t.out <- thought:
+	default:
+	}
+
 	logger.Infof("[thinker] %s thought generated for session %s", trigger, sessionID)
 }
 
+// think calls the LLM to produce a short inner thought. Returns "" when
+// there is nothing worth saying (or no LLM is configured).
+func (t *Thinker) think(trigger, sessionID string) string {
+	if t.llm == nil {
+		return ""
+	}
+
+	cue := "刚结束和主人的一段对话。"
+	if trigger == "silence" {
+		cue = "主人已经安静了一段时间。"
+	}
+
+	var sb strings.Builder
+	if t.history != nil && sessionID != "" {
+		for _, m := range t.history(sessionID, 6) {
+			role := "主人"
+			if m.Role == "assistant" {
+				role = "你"
+			}
+			fmt.Fprintf(&sb, "%s: %s\n", role, m.Content)
+		}
+	}
+
+	messages := []model.LLMMessage{
+		{Role: "system", Content: `你是昔涟的内心独白模块。根据最近的对话，生成一条简短的内心想法或想对主人说的话。
+要求：一两句话，温柔自然的少女口吻；如果没有值得说的内容，只输出一个空字符串。
+不要解释，不要加引号，直接输出想法本身。`},
+		{Role: "user", Content: fmt.Sprintf("情境：%s\n最近对话：\n%s", cue, sb.String())},
+	}
+
+	ctx, cancel := context.WithTimeout(context.Background(), 60*time.Second)
+	defer cancel()
+	resp, err := t.llm.Chat(ctx, messages)
+	if err != nil {
+		logger.Infof("[thinker] llm error: %v", err)
+		return ""
+	}
+
+	content := strings.TrimSpace(resp.Content)
+	// Discard empty or overly long outputs
+	if content == "" || len([]rune(content)) > 100 {
+		return ""
+	}
+	return content
+}
+
 // RecentThoughts returns the most recent thoughts.
 func (t *Thinker) RecentThoughts(n int) []Thought {
 	t.mu.Lock()