A vision-language model and platform for temporally mapping surgery from video