I don't know the LessWrong-like answer, so I can only offer you the human, empathic answer.

Based on the phrasing of your question:

whether someone born a male but who identifies as female is indeed female

and the fact that you have posted it to LessWrong, I understand it to be a question about constructing a useful and consistent model of the human condition, rather than about respecting an actual or hypothetical human being. If so, I think you are asking the wrong question.

Your students want to learn from you, but on a more basic level, they want to feel safe with you. If you have a trans student, or a student with a trans friend/relative, she is likely to take your answer to this question very personally. Your choice boils down to whether you offer a personal welcome (by recognizing your student's identity) or a personal affront (by implying that you have more authority than she does to determine who she "really is").

I should add that it is a common failure mode for humans, when confronted with a counterexample to their existing model of the human condition, to insist that their model is correct and that the fellow human they are dealing with is a bad data point. As well as rude and demeaning, this approach is irrational and intellectually dishonest.