Under dataflow RunnerV2, after trySplit returns successfully, the runner does not immediately check the progress accounted for by testing the RestrictionTracker. Instead, it destroys the RestrictionTracker and tries to construct a new one using the already claimed restriction instead. The runner will not do this if the SDF implements the @Getsize annotated method, in which case it will be used instead. As far as I can tell, this particular interplay is undocumented, but it should remain true for this public API.

Our I/O will crash when constructed with an already completed restriction to avoid constructing a restriction tracker in an unusable state. We should implement GetSize so that we don't need to allow for partially unusable RestrictionTrackers.