But it does. How many grains of send form a heap? How many atoms form a fluid? Is it something a particle or a wave? The line can be drawn depending on specific problem, on how useful a heap/fluid/particle/wave model for this specific application.
A label (abstraction) allows us to bring corresponding tools that were developed for it. If you can count trees then the same math can be useful to count people.
Arguing about where you draw the line doesn't advance anything, it just muddies the water. We have robust definition of what is and isn't alive built by consensus. Mitochondria fall into the 'not alive' category by our definition. Presenting decades old scientific views as evidence that this categorisation is wrong doesn't add anything. This article has had far more exposure than some really groundbreaking science and it adds nothing.
My point is more general: there may be multiple useful descriptions of “reality” depending on context. It is almost _trivially_ (tautologically) true. Ask two different programmers to implement something and look at what abstractions they create depending on unrelated non-functional requirements. Here’s specific example: imagine you are writing “what-if” type of article about a table: if you are interested in whether it holds your weight then you might talk in terms of tension, compression, mechanical forces. If you are interested in whether you can hide behind it from x-rays then terms such as radiolucent, atomic composition might be useful.
I have no idea in what context “mitochondria are alive” notion might be useful (but it doesn’t mean there is none).
Defining an arbitrary line and then attaching labels does not really add to understanding.