Key features:
Latency-free AI: uses HuggingFace's Candle framework for inference within the Rust process (no Python overhead).
Memory-optimized: uses Slab allocators for high-frequency agent management.
Detects spoofing, layering, and wash trading in real time.