Google’s TurboQuant AI-compression algorithm can reduce LLM memory usage by 6x https://arstechnica.com/ai/2026/03/google-says-new-turboquant-compression-can-lower-ai-memory-usage-...