diabetes: one-hot columns of one variable are acquired one at a time #145

Problem

diabetes holds one-hot columns from Kachuee et al.'s preprocessing: race and ethnicity (RIDRETH3, RIDRETH1, columns 2–12) and smoking (SMQ020, columns 38–41). Each column is a separate feature with the direct unmasker, so a method can acquire "is the participant in race category 3" without the rest of the variable, and pays per column. In reality one question answers the whole variable.

Scope

• Dataset generation and the unmasker for diabetes; both budget settings. Not in the kdd26 dataset list.
• Depends on Rebuild diabetes from NHANES, and decide what to do about its imputed labels #137, the diabetes rebuild, which may change the columns.

Options to decide

• Group each variable's one-hot columns into one feature group (needs an unmasker for tabular groups, as Nominal categorical features are label-encoded as ordinal integers, inconsistently across datasets #143 may also need).
• Or replace the one-hot columns with one categorical feature each.

Acceptance criteria

• Acquiring a diabetes variable reveals the whole variable, or the page documents why not.
• just qa passes.