Step 3, that a reason to doubt reliability is a reason to doubt everything, is too strong. We can doubt our faculties in abstract metaphysics while trusting them in perception and survival-relevant reasoning, where selection pressure clearly applied.

This is one of the better responses in the literature. The reply is that the argument for naturalism itself is abstract metaphysics, so the domain restriction would undermine naturalism specifically while leaving everyday reasoning intact, which is close to the conclusion Plantinga wants. Whether faculties can be trusted domain by domain in this way is a fair question, and it is one you should weigh rather than take on trust from either side.