Moved to Q&A.
@balrog and
@Mack are correct.
To further answer your question, the dispatching time and block length will also have been designed in such a way so as to mean the scenario of needing to hold a train on a block in a "non-emergency" situation is very unlikely.
If you imagine Helix as an example, there will be a sensor at the end of the second launch that confirms the train has reached the correct launch speed (due to the way LSMs work, this information can be determined by the launch system itself). Only after this has been confirmed*, will the first launch be 'ready'. If this doesn't happen, then the first launch will go into E-stop mode, behaving exactly as Mack has described above.
*Sometimes there is also a sensor placed just after the high point of the coaster (so in Helix's case, just after the Top Hat after the second launch). This sensor, again as Mack and Balrog have described above, is designed to tell the controls that the train has cleared the high point. Logic then dictates that the train can't get back past this point, and so it's safe for another train to enter the first launch.
By using both of these systems (the proving sensors and the layout design), regardless of a rollback or stall, no two trains are ever in the same block.