Incorrect.
The overall poll results are based on the sample of 800, including whatever Latinos were caught in the "random" net.
In addition, they also did a supplementary sample of 125 Latino "likely voters", which were added to the Latinos within the 800 to create a separate sub-sample of Latino voters for separate analysis.
That is a legit polling practice, as it creates a Latino sample large enough to be sliced and diced for analytical purposes.