#124 adds MLX support. We need to adopt the [in-place KV cache update](https://github.com/ml-explore/mlx-swift-examples/blob/main/Libraries/LLM/KVCache.swift#L9) pattern to improve performance.