I wouldn't say the last step that you attribute to me. Firstly, if I were going to talk about the long run, I would say that in the long run, you should maximise expected utility because you'll probably get a lot of utility that way. That being said, I don't want to talk about the long run at all, because we don't make decisions for the long run. For instance, you could decide to have a bacon omelette for dinner today and then stay veg*n for the rest of your life, and the argument that you attribute to me wouldn't work in that case, although I would urge you to not eat the bacon omelette. (In addition, the line of reasoning that I would actually want you to use would involve attributing >50% probability of cow, chicken, pig, sheep, and fish sentience, but that's beside the point).

Rather, I would make a case like this: when you make a choice under uncertainty, you have a whole bunch of possible outcomes that could happen after the choice is made. Some of these outcomes will be better when you choose one option, and some will be better when you choose another. So, we have to weigh up which outcomes we care about to decide which choice is better. I claim that you should weigh each outcome in proportion to your probability of it occurring, and the difference in utility that the choice makes. Therefore, even if you only assign the "cows are sentient" or "theory X is true" outcomes a probability of 20%, the bad outcomes are so bad that we shouldn't risk them. The fact that you assign probability >50% to no damage happening isn't a suffcient condition to establish "taking the risk is OK".