Balancing Performance and Consistency in Pandas Online Usage: Optimizing DataFrame Processing for Machine Learning Pipelines