Optimize Cardinality Estimation Model Pretraining by Simplifying the Training Datasets