Describir: Autoencoders for Financial Anomaly Detection: A Review of Architectures, Evaluation Practice and Deployment Gaps