Yes, I know and used tune with Bayesian optimization but in this particular case it wouldn't work I think (perhaps I should have elaborated more). I'm working with LightGbm with a particular time-series data structure and I don't think tune/ caret can be flexibly used in such case without converting the model to a parsnip specific format right?
On the other hand Optuna is generic/ framework agnostic - do you know of anything like that for R?