Add "local" or "nearest neighbors" imputation for missing visits

[gabrielodom]

If we see a "-o+" use pattern, then the "o" probably should not be imputed to "-" (as would happen with LOCF or LOCF-D); it should be "+". We'd need to add this as a new method to impute_missing_visits() and include other arguments (probably through ...).