Our household has given this a lot of thought. Our humble proposal is a hybrid-model. Consider:

Reviews are anonymous throughout the review period. All of your points hold here. If the paper is rejected the reviews remain anonymous forever. If the paper is accepted then the positive reviews become public, and shared as part of the Sup Info for the paper. Any negative reviews remain anonymous; we go back and forth on whether they should be Sup Info too. The author of the positive review has the option of including that effort in a "Reviews" section of his/her CV, which does not carry as much weight as the rest of the CV but at least allows that this was a non-trivial effort executed with professional attention.