There exist plenty of definitions that can be imagined which make the above snowflakes very rare in practice.
If the beholder argument is that the existence of a single such snowflake invalidates the possibility that useful objectivity can be used to study the concept of trolling -- then the claim is ludicrous. We can study the notion of trolling without getting every single label right every time.
Agreed, but I don't think we're talking about snowflakes. I think we're talking about a significant portion of the communication out there that might be identified by some as "trolling". And I guess I'm less interested in an objective definition of trolling than I am in an objective determination of whether some particular post or group of posts is intended as a troll.
On top of that, I don't think it's a binary determination. There's such a thing as a 100% troll post or a 100% innocent post, but there's also everything in between.
I took it more as an expression of something like Poe's Law. If someone says something inflammatory, is it a troll, or is it an honest expression of opinion? You can theorize about the author's mental state, but you can't know what they're thinking. Without extra information, the difference rests in the reader's perception.