Good questions!
Short answer:
Yes, Category 2 is correct.
Transcribe all t-pages and transcriber notes in contracted UEB. On the Special Symbols page, list UEB symbols and symbols from the other language’s code under separate headings. (For a language that uses a modified Latin alphabet, letters that are the same in UEB and the other language do not need to be listed on the SS page, but other alphabetic symbols like modified letters or tone markings should be included.)
Long answer:
BANA is not explicit about every aspect of this type of transcription, but I take the general assumption to be that a braille reader in North America will have UEB as their primary code. This reader may have previous experience with another code while living in a different country, but is now in North America, where UEB is the default. Or they may be a speaker of another language without ever learning the full braille code, or may simply be multi-lingual and reading text in a non-English language in addition to being a reader of English text in UEB. It is safer, as a rule, to assume that UEB is the primary code for someone in North America. Based on this assumption, Category 2 is the best choice for a fully non-English book unless specifically requested otherwise.
The guidelines specify that the title page is in UEB. It is logical that all transcriber-generated content should be in a single code (UEB) so that the reader knows which code they are reading. My advice is to do all the t-pages, and all transcriber’s notes, in English. This is not explicit in the guidelines, but I think it is what makes the most sense; the reader should not have to guess at which language is being used when, and transcribers are (as a general rule) not qualified to be translators and should avoid taking on that role.