[CUDA] Tune ops per buffer based on device#2761

awni merged 3 commits intoml-explore:mainml-explore/mlx:mainfrom

awni:tune_ops_per_bufferawni/mlx:tune_ops_per_bufferCopy head branch name to clipboard