have `simulate()` validate output by checking for negative values or NaNs

[papsti]

negatives can get lost in cases where aggregation over age groups and/or epi state subcategories occur; would be nice to have simulate() throw an error before any further data manipulation occurs.