feat: Antigravity extra failover retries after default retries exhausted

When default failover retries are exhausted, continue retrying with
Antigravity accounts only (up to 10 times, configurable via
GATEWAY_ANTIGRAVITY_EXTRA_RETRIES). Each extra retry uses a fixed
500ms delay. Non-Antigravity accounts are skipped during the extra
retry phase. Applied to all three endpoints: Gemini compat, Claude,
and Gemini native API paths.

backend/internal/handler/gemini_v1beta_handler.go

@@ -323,6 +323,7 @@ func (h *GatewayHandler) GeminiV1BetaModels(c *gin.Context) {
 
 	maxAccountSwitches := h.maxAccountSwitchesGemini
 	switchCount := 0
+	antigravityExtraCount := 0
 	failedAccountIDs := make(map[int64]struct{})
 	var lastFailoverErr *service.UpstreamFailoverError
 	var forceCacheBilling bool // 粘性会话切换时的缓存计费标记
@@ -340,6 +341,15 @@ func (h *GatewayHandler) GeminiV1BetaModels(c *gin.Context) {
 		account := selection.Account
 		setOpsSelectedAccount(c, account.ID)
 
+		// 额外重试阶段：跳过非 Antigravity 账号
+		if switchCount >= maxAccountSwitches && account.Platform != service.PlatformAntigravity {
+			failedAccountIDs[account.ID] = struct{}{}
+			if selection.Acquired && selection.ReleaseFunc != nil {
+				selection.ReleaseFunc()
+			}
+			continue
+		}
+
 		// 检测账号切换：如果粘性会话绑定的账号与当前选择的账号不同，清除 thoughtSignature
 		// 注意：Gemini 原生 API 的 thoughtSignature 与具体上游账号强相关；跨账号透传会导致 400。
 		if sessionBoundAccountID > 0 && sessionBoundAccountID != account.ID {
@@ -424,15 +434,23 @@ func (h *GatewayHandler) GeminiV1BetaModels(c *gin.Context) {
 			var failoverErr *service.UpstreamFailoverError
 			if errors.As(err, &failoverErr) {
 				failedAccountIDs[account.ID] = struct{}{}
+				lastFailoverErr = failoverErr
 				if needForceCacheBilling(hasBoundSession, failoverErr) {
 					forceCacheBilling = true
 				}
 				if switchCount >= maxAccountSwitches {
-					lastFailoverErr = failoverErr
-					h.handleGeminiFailoverExhausted(c, lastFailoverErr)
-					return
+					// 默认重试用完，进入 Antigravity 额外重试
+					antigravityExtraCount++
+					if antigravityExtraCount > h.antigravityExtraRetries {
+						h.handleGeminiFailoverExhausted(c, failoverErr)
+						return
+					}
+					log.Printf("Gemini account %d: antigravity extra retry %d/%d", account.ID, antigravityExtraCount, h.antigravityExtraRetries)
+					if !sleepFixedDelay(c.Request.Context(), antigravityExtraRetryDelay) {
+						return
+					}
+					continue
 				}
-				lastFailoverErr = failoverErr
 				switchCount++
 				log.Printf("Gemini account %d: upstream error %d, switching account %d/%d", account.ID, failoverErr.StatusCode, switchCount, maxAccountSwitches)
 				if account.Platform == service.PlatformAntigravity {