A deeper look at Llama 4 — Scout and Maverick, the 10-million-token context window, the mixture-of-experts architecture and what the release changes for people building on open models.