Describir: Bridging the Semantic Gap: An Ensemble Learning Framework With Textual Topic‐Raw Financial Feature Fusion to Enhance Fraud Detection in Chinese Markets