feat(gateway): 实现负载感知的账号调度优化

- 新增调度配置：粘性会话排队、兜底排队、负载计算、槽位清理
- 实现账号级等待队列和批量负载查询（Redis Lua 脚本）
- 三层选择策略：粘性会话优先 → 负载感知选择 → 兜底排队
- 后台定期清理过期槽位，防止资源泄漏
- 集成到所有网关处理器（Claude/Gemini/OpenAI）

backend/internal/handler/gateway_helper.go

@@ -83,6 +83,16 @@ func (h *ConcurrencyHelper) DecrementWaitCount(ctx context.Context, userID int64
 	h.concurrencyService.DecrementWaitCount(ctx, userID)
 }
 
+// IncrementAccountWaitCount increments the wait count for an account
+func (h *ConcurrencyHelper) IncrementAccountWaitCount(ctx context.Context, accountID int64, maxWait int) (bool, error) {
+	return h.concurrencyService.IncrementAccountWaitCount(ctx, accountID, maxWait)
+}
+
+// DecrementAccountWaitCount decrements the wait count for an account
+func (h *ConcurrencyHelper) DecrementAccountWaitCount(ctx context.Context, accountID int64) {
+	h.concurrencyService.DecrementAccountWaitCount(ctx, accountID)
+}
+
 // AcquireUserSlotWithWait acquires a user concurrency slot, waiting if necessary.
 // For streaming requests, sends ping events during the wait.
 // streamStarted is updated if streaming response has begun.
@@ -126,7 +136,12 @@ func (h *ConcurrencyHelper) AcquireAccountSlotWithWait(c *gin.Context, accountID
 // waitForSlotWithPing waits for a concurrency slot, sending ping events for streaming requests.
 // streamStarted pointer is updated when streaming begins (for proper error handling by caller).
 func (h *ConcurrencyHelper) waitForSlotWithPing(c *gin.Context, slotType string, id int64, maxConcurrency int, isStream bool, streamStarted *bool) (func(), error) {
-	ctx, cancel := context.WithTimeout(c.Request.Context(), maxConcurrencyWait)
+	return h.waitForSlotWithPingTimeout(c, slotType, id, maxConcurrency, maxConcurrencyWait, isStream, streamStarted)
+}
+
+// waitForSlotWithPingTimeout waits for a concurrency slot with a custom timeout.
+func (h *ConcurrencyHelper) waitForSlotWithPingTimeout(c *gin.Context, slotType string, id int64, maxConcurrency int, timeout time.Duration, isStream bool, streamStarted *bool) (func(), error) {
+	ctx, cancel := context.WithTimeout(c.Request.Context(), timeout)
 	defer cancel()
 
 	// Determine if ping is needed (streaming + ping format defined)
@@ -200,6 +215,11 @@ func (h *ConcurrencyHelper) waitForSlotWithPing(c *gin.Context, slotType string,
 	}
 }
 
+// AcquireAccountSlotWithWaitTimeout acquires an account slot with a custom timeout (keeps SSE ping).
+func (h *ConcurrencyHelper) AcquireAccountSlotWithWaitTimeout(c *gin.Context, accountID int64, maxConcurrency int, timeout time.Duration, isStream bool, streamStarted *bool) (func(), error) {
+	return h.waitForSlotWithPingTimeout(c, "account", accountID, maxConcurrency, timeout, isStream, streamStarted)
+}
+
 // nextBackoff 计算下一次退避时间
 // 性能优化：使用指数退避 + 随机抖动，避免惊群效应
 // current: 当前退避时间