It previously had to iterate over EVERY task in its list every time. With lots of queued tasks that is slow.
Instead we sort it by schedule time. That means when iterating thru list, if we get to a single task that is in the future, so are all remaining.
In addition to fixing the problem, it makes sense. The observer pattern is overkill for an object type we own; doing it directly in Destroy() is more efficient.
* If the "scheduler" controller is restarted by the process scheduler for
any reason, it will start throwing runtimes every status panel update
because the new replacement instance doesn't initialize its list of
schedule tasks.
* Fix that by copying over the unfinished list from the old instance, but
doing some safety checks to make sure it doesn't copy over bad stuff.
