Prediction of TB treatment outcomes among HIV/TB coinfected patients in Uganda using routinely collected clinical data: a machine-learning approach