fix(openai): persist passthrough 429 rate limits

backend/internal/service/openai_gateway_service.go

@@ -2598,6 +2598,12 @@ func (s *OpenAIGatewayService) handleErrorResponsePassthrough(
 	}
 	setOpsUpstreamError(c, resp.StatusCode, upstreamMsg, upstreamDetail)
 	logOpenAIInstructionsRequiredDebug(ctx, c, account, resp.StatusCode, upstreamMsg, requestBody, body)
+	if s.rateLimitService != nil {
+		// Passthrough mode preserves the raw upstream error response, but runtime
+		// account state still needs to be updated so sticky routing can stop
+		// reusing a freshly rate-limited account.
+		_ = s.rateLimitService.HandleUpstreamError(ctx, account, resp.StatusCode, resp.Header, body)
+	}
 	appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 		Platform:             account.Platform,
 		AccountID:            account.ID,