Intent-to-treat analysis of cluster randomized trials when clusters report unidentifiable outcome proportions