The table is correct. First because design intent is irrelevant. Second because the author clearly states that the focus of his comparative study is on *-preimage resistance vs collision. There is more to the article than the table :)
I really do not understand what you are trying to say. Just like SPHINCS, Ed25519 only cares that the hash function used with it has *-preimage resistance and it would not be broken if a collision was found in it. See https://ed25519.cr.yp.to/ed25519-20110926.pdf
This table is incorrect. Ed25519 was explicitly designed with the intention of only needing preimage resistance, not collision resistance.