The point is just that factor analysis assumes that the items/​variables end up correlating due to the factors. If you put variables that are upstream of the factors, such as sex, into the factor analysis, then those upstream variables would have no reason to correlate with each other in ways that match the factor structure (and in fact due to collider bias, would in this case have reasons to end up correlated in ways that precisely oppose the factor structure), so therefore it would be nicer to avoid demographic variables as much as possible.
The point is just that factor analysis assumes that the items/​variables end up correlating due to the factors. If you put variables that are upstream of the factors, such as sex, into the factor analysis, then those upstream variables would have no reason to correlate with each other in ways that match the factor structure (and in fact due to collider bias, would in this case have reasons to end up correlated in ways that precisely oppose the factor structure), so therefore it would be nicer to avoid demographic variables as much as possible.