How did someone find its existence?
None.
I'm guessing a combination of technical knowledge and a bit of luck.
An artist's depiction of an Extended Unit Death
I would have discovered them on my own if someone else hadn't. I'm fairly confident any mapping enthusiast would have discovered them independently.
It's all about observations.
What happens when you preserve a trigger? Well, the trigger runs, then after about a 2-second delay, it repeats. Pretty basic stuff. But after a while, when you're adding more triggers to your map, you'll notice the delay on your preserve triggers become inconsistent; sometimes they're fast, and sometimes they're slow. Is it changing because you have more triggers? Making a bunch of Create/Kill Unit triggers doesn't seem to affect it, so that can't be the answer. What trigger(s) did you add that caused this? You can start deleting triggers until the inconsistency goes away, and then look at the trigger(s) you deleted to see what they contained that might influence the delay. Well, the "Wait" action looks mighty suspicious right now.
If you notice that your preserved triggers run faster when you have a trigger with waits executing, it should get you to ask questions about why that is. If you decrease the amount of time on the waits for that trigger, the other triggers also speed up. So, what if we have a trigger that just spams Wait 0? Well, all the other triggers run insanely fast! But there's a problem here: they only do it for a short while, and then there's the regular long delay. Well, how many times to they run quickly before the delay? In the case of one hyper trigger, you could count it out to be up to 63 times, or the number of waits you used. Naturally, if you want the effect to last longer, you would just copy that trigger. But how many waits do you need?
From here, assuming you don't already know a lot on trigger execution, you could copy the trigger once and see how much more time it provides you. I would have probably expected it to only double, since we're doubling the waits, but what actually happens is quite different: we have the first trigger run 63 times and the second trigger executes one of its waits, but then the first trigger starts over again. From this, we can assert that the amount of time given to us is multiplicative. Well, copying the trigger four times would give us 15752961 trigger cycles before a long pause. Assuming about 12 cycles per second, this would give us a little more than 364 hours (about 15 days) before all those triggers would cycle through and we'd get our long delay again.
That's how I would have discovered them, at least; I imagine the actual discovery was through a similar process.