FWIW, I recently came across the SurvMetrics R package that implements several prediction accuracy metrics for time-to-event/survival analysis.
Perhaps it would be possible to incorporate some of these metrics into the tidymodels survival framework?