Fix: Force BASE_MODEL and correct fraud_probability variable in frontend

Fix: Add robust error handling and frontend ERROR state

fix: Reverter motor para a versão estável pré-motor pessoal (v2.7)

Rollback: Returning to stable HyperMoon model as primary engine v2.9.4

Fix: Improved detection sensitivity and removed quantization bias v2.9.3

Deployment: Synchronized system with new model architecture and streaming notebook v3.0

Comprehensive Refinement Cycle v2.6: Security, Performance, and Training Quality

Support custom model repository via CUSTOM_MODEL_REPO env var

ML Upgrade: HyperMoon sovereignty, dynamic quantization, and batch processing v2.4

Refactor: Admin & Result UI consistency, glassmorphism polish, and backend sync

Cleanup: Repositório otimizado (código + dashboard apenas)