We need to test the error propagation behavior when a multi-shard update or findAndModify fails because of rejection by the ingress operation rate limiter. We should cover both the budget exhaustion and retry cap scenarios.