There's nothing suspicious here. NYT has consistently reported result among both registered voters and likely voters in their crosstabs. They use a likely voter model where every registered voters gets some weight, so the sample size among all registered voters is relevant.