Conventional wisdom has always been that pures need 4 ranks of armor use to train off the action penalty for full leathers. But I was looking at the charts and fulls have an AP of 0. Does that mean they have a nonzero AP that's trained down to 0 at 4 ranks?

The chart shows the penalty at normal training. With less training there is a higher penalty, and with substantial overtraining a non-zero penalty can be reduced.