The proof the author has given is equivalent to building a Turing machine out of existing programs in his language of interest. M:tG cards have instructions printed on them in the ability text language - so those instructions, rather than the cards themselves are the symobls in the language. The cards are best seen as scripts or small programs written in that language.
But why should anyone restrict themselves to existing cards when trying to prove the Turing completeness of the language?
Note, also, that the rules of well-formed sentences in the ability text language are given in the M:tG Comprehensive Rules or at least some of them are- there's no complete formal definition of the language. Still, there's plenty enough for anyone to be able to form syntactically correct ability text.
So, again- if you're trying to prove the Turing completeness of a language, why would you not use the full expressive power of the language to compose new strings? And instead restrict yourself to just a few pre-existing strings of it?
BTW, the kind of language the ability text language is, has a name: it's a Controlled Natural Language. [https://en.m.wikipedia.org/wiki/Controlled_natural_language].
As to the relation between the language and the rules, if you think about it, if the language is not equivalent to the rules, then it can't represent (and, therefore, modify) every rule in the game (as it is supposed to). And if the language and the rules are equivalent, then proving one is Turing complete also proves it for the other. So you can ignore the rules and focus any proof on the language.
Given such a definition of the ability text language, you don't need to restrict yourself to only those strings that are actually printed on cards.
In the same sense, we don't define the English language as every text ever written in English - but as some unknown formal system that can produce an unknown, probably infinite, number of strings that includes all text ever written in English.
So why treat the M:tg language any different?
I think you might be confused by the term "language". In the context of the proof, "language" means "anything that is the Oracle text of a card". You seem to be using it as "anything that could be printed on a card". I think you're saying "the game of Magic-but-where-I'm-allowed-to-choose-what-the-cards-say is Turing complete", which is undeniably true but also not the point.