Yes, good question. There are relay standardness policies like minrelayfee that are about DOS protection for nodes in terms of bandwidth etc.
I should probably research it, but as i remember there are others that are not like that, perhaps non-standard scriptpubkey types, and those would fit into the rubric of what I'm saying here, agreed.
Hmm as I write that I realized the obvious point that unrestricted scriptpubkey types might themselves represent a dos threat: but there's a still a distinction to be made, right, between limits on size (and parsing computation), and limits on "type".
There is a distinction and it indeed gets a bit fuzzy. I.e. outputs bigger than 64?bytes (iirc) require extra allocations when writing to disk, but obviously that's no argument for prohibiting other types. A weaker reason with stronger (lol!) assumptions about its usage is that it should be theoretically spendable unless marked as such. Some script pubkeys might not be, which is a reason to only allow "types" that at least formally are.
Yeah i came to the same conclusion when i thought about it. It's obviously the practical choice to use a limited set, while the op_return carve out for data is ... not obvious, but has a clear purpose but can be argued against, for sure.
