Enabling Composable Scalable Memory for AI Inference with CXL Switch | Open Compute Project | Podwise