DEPT edits by multiplicity, which a proton-decoupled 13C spectrum cannot:
135 puts CH and CH3 positive, CH2 NEGATIVE, and drops quaternaries entirely;
90 shows only CH. That is what makes it worth running, and why this is a
separate tool from covnmr_predict_spectrum rather than a nucleus option.

Each returned peak carries `sign` (+1 / -1) and `dept_class` (CH, CH2, CH3),
and the rendered curve carries the sign too -- there is no `intensity` field
on a DEPT peak, the sign is the information.