A query on an Azure SQL Database joins a large orders table to a filtered set of customers. The number of qualifying customer rows varies dramatically between executions depending on the WHERE predicate values. You observe that some executions use a nested loops join (fast for small inputs) while others suffer badly because a nested loops plan was cached and reused when the input was actually large. The database is on a recent compatibility level. Which Intelligent Query Processing feature will let the engine defer the choice of physical join type until the actual number of rows is known at runtime?