Yes, that’s correct - aside possibly from the FACT algorithm, where the cutoff presumably refers to the length of the vector (@rsmith will no doubt correct me if I’m wrong).
With deterministic tracking, it is used for termination. For probabilistic tracking, it serves both functions you suggest, but the latter is the more correct way of phrasing it: directions whose FOD amplitude is lower than the cutoff are not considered - they are effectively removed from the PDF being sampled. When no direction above the cutoff can be found within the allowed solid angle, then the streamline is terminated.
Cheers,
Donald.