fix(feishu): group pairing uses group-level ID instead of per-user

Changed Lark group pairing to use "group:<chatID>" as sender_id
(matching Telegram pattern) so one approval covers the entire group.
Added approvedGroups in-memory cache to avoid DB queries per message.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

internal/channels/feishu/bot_policy.go

@@ -66,10 +66,7 @@ func (c *Channel) checkGroupPolicy(senderID, chatID string) bool {
 		}
 		return false
 	case "pairing":
-		paired := false
-		if c.pairingService != nil {
-			paired = c.pairingService.IsPaired(senderID, c.Name())
-		}
+		// Allowlist bypass (per-user)
 		inAllowList := c.HasAllowList() && c.IsAllowed(senderID)
 		inGroupAllowList := false
 		for _, allowed := range c.groupAllowList {
@@ -78,12 +75,21 @@ func (c *Channel) checkGroupPolicy(senderID, chatID string) bool {
 				break
 			}
 		}
-
-		if paired || inAllowList || inGroupAllowList {
+		if inAllowList || inGroupAllowList {
 			return true
 		}
 
-		c.sendPairingReply(senderID, chatID)
+		// Group-level pairing (one approval per group, matching Telegram pattern)
+		if _, cached := c.approvedGroups.Load(chatID); cached {
+			return true
+		}
+		groupSenderID := fmt.Sprintf("group:%s", chatID)
+		if c.pairingService != nil && c.pairingService.IsPaired(groupSenderID, c.Name()) {
+			c.approvedGroups.Store(chatID, true)
+			return true
+		}
+
+		c.sendPairingReply(groupSenderID, chatID)
 		return false
 	default: // "open"
 		return true

internal/channels/feishu/feishu.go

@@ -39,6 +39,7 @@ type Channel struct {
 	dedup           sync.Map // message_id → struct{}
 	pairingDebounce sync.Map // senderID → time.Time
 	reactions       sync.Map // chatID → *reactionState
+	approvedGroups sync.Map // chatID → true (in-memory cache for paired groups)
 	groupAllowList  []string
 	groupHistory    *channels.PendingHistory
 	historyLimit    int