We should implement a 'hit rate' top-N metric that measures whether the recommendation list contains any test items.

When user test sets have size 1, this is equivalent to recall, but having a specific metric (1) makes it easier to use and (2) handles larger test set sizes.