In the code, I couldn’t find any 2^64 bound on the lexeme count, so removing that makes sense. The added sentence about distinct lexeme count seems to overlap with the existing description of tsvector limits, so I’m not sure it adds much new information.