Llama 4 Inference Fixes, DeepSeek's Silence, and New Data Tooling

Llama 4 Engine Fix Restores vLLM Performance

The vLLM team squashed a critical performance regression in the Llama 4 Eagle speculative decoding engine. If you're running large models in production, this patch fixes the engine so you actually get the speedup you were promised. It's the kind of boring but essential infrastructure work that keeps your stack stable.