Consider waiting for child processes before exit

In the case that the cmd starts child processes and exits before the children, duct will exit when the parent exits.

It might be good (though I am a little worried about adding complexity) to keep track of all the children pids, and wait for them to stop before we stop measuring.

Start by tracing how duct launches the command, handles child processes, and decides when measurement stops; the issue does not name specific files or tests. Define how child PIDs are tracked and what waiting means for process termination, then verify that duct does not stop measuring until the relevant children have stopped.

Written by the indexing model from the issue text.