[vcnainala]

Summary

In the per-atom quality report of validate-assignments, the two protons of a diastereotopic CH2 were relabelled by stripping the last character of the author's label and appending a/b. Labels that do not end in a/b, such as Mnova's H-5'ax / H-5'eq, came out mangled, and the pair field pointed at a label that does not exist.

Both rows now keep the label the author gave each shift, and pair names the other proton's real label. Ordering is unchanged: the lower shift comes first.

labelForAtom and the new rowLabelForAtom share the per-atom label lookup, so multi-atom rows (H-2, H-6) still split into their per-atom names.

Follow-up to #147.

Test plan

• npm test in app/scripts/nmr-cli: 13 of 13 pass, including a new nicotine case where H-5'a/H-5'b are renamed to H-5'ax/H-5'eq and both the row labels and the pair references are checked.