A data engineering team uses Azure Databricks with the Azure Cosmos DB Spark 3 connector to load a large historical dataset into a provisioned throughput container that is also serving a production read workload. During the initial bulk write job, the operations team observes that production queries experience elevated latency and 429 (rate limited) responses. The team wants the Spark job to consume only a bounded fraction of the container's provisioned RU/s while still completing, without permanently increasing the container's throughput. Which configuration should they apply to the Spark write?