Previously in order to wait on a job object we would create an IO
Completion Port, configure the job object to emit notifications to it
with SetInformationJobObject, and wait for JOB_OBJECT_MSG_EXIT_PROCESS
notifications until all processes have died.

This followed one piece of guidance from Microsoft [1] but according to
Microsoft's own documentation, this cannot work reliably as delivery of
job notifications is not guaranteed [2]. I have seen cases where the
processes hang waiting on job objects so I can only guess that message
loss is indeed possible.

Instead we now take a simpler approach: look at the processes in the
job, if there are none then we are done. If there are still processes,
choose one and wait for it to finish. Iterate until the job is empty.

Credit for this approach goes to Davean Scies.

[1] https://devblogs.microsoft.com/oldnewthing/20130405-00/?p=4743
[2] https://docs.microsoft.com/en-us/windows/win32/api/winnt/ns-winnt-jobobject_associate_completion_port