How should duplicate detection be handled during a large-scale initial data load into Salesforce when Duplicate Rules may create performance bottlenecks at load time?