Using abort_callback from ggml to interrupt llama computation

Using abort_callback from ggml to interrupt llama computation #2328