The point is that NIP-94 wasn't meant to be for embedded images in kind:1 notes. If someone really wants to reference a kind:1063 they should be able to, paste a nevent code and that's fine, but they shouldn't expect other social clients to be able to understand these events, so it makes no sense to make that the default way of loading images.

The best way to ensure that the protocol is kept simple is to not assume other clients will implement whatever NIP you think it's cool -- you can implement anything you want, but you should still make it so the experience works for others that don't.