BUG: AzureContentFilterScorer.get_scorer_metrics always returns None #3544

AzureContentFilterScorer only sets evaluation_file_mapping inside evaluate_async. A scorer you just built for one category has no mapping, so get_scorer_metrics() returns None even though the registry has its entry with a matching eval hash.

scorer = AzureContentFilterScorer(harm_categories=[TextCategory.HATE])
scorer.evaluation_file_mapping   # None
scorer.get_scorer_metrics()      # None, but hate_speech_metrics.jsonl has this eval hash

Same for violence, sexual and self harm. Setting the mapping in the constructor for single-category scorers fixes it, like the Likert scorers already do.