Your second and third options are actually the same. In general, you want something of the form y + (S-y-n) * p, where p is your best guess or estimate of the proportion of people who have not RSVPed who will end up deciding to come. Your first option is this with p = 1/2, and your second option is this with p = the empirical ratio of yes and no so far. The correct choice of p is not directly a question that math can solve, because it will depend heavily on context and what other information you have. What actually *is* your best guess, taking into account all the information you've seen?
If you have a strong belief (based on past experience, context, external knowledge, or any other such reason) that the remaining people who haven't responded are actually 50-50 to come despite the empirical ratio so far, you should choose the first option with p = 1/2.
If you are very unsure and believe that the empirical ratio of yes and no in responses so far is your best estimate for p, you should choose the second option with p = the empirical ratio so far.
If for various reasons (past experience, context, etc.) you believe that people who haven't responded up until now are very unlikely to come, you should choose p to be a very small value, matching your belief.
If you are unsure and believe that the empirical ratio is a good estimate, except that maybe people who haven't responded are slightly less likely to come because putting it off indicates that they aren't as interested, you should choose p = the empirical ratio minus a little bit.
Also, the choice of how much food to buy probably should depend not just on p and the expected number of people, but also on the relative costs of having too much and having too little food, which again depends on context. Between buying too much food and buying too little food, is one significantly worse than the other?