First of all thanks for this library. It used some macro's i haven't seen before.
I use library's like this on atmel 328. Those have 1024 bytes of ram. So i always try to limit the memory usage to as little as possible.

So i have some questions as i do not fully understand the bidirectional registry.
U can read and write only one byte to the register. If u write one byte u would read the same byte back. Because it would set a low powered pullup or down it's a in and output at the same time.

At startup all pins are high. Given no pins are grounded it will give 0xFF.
So i only think u need two bytes or memory. One for (last) readed value. One for last written value. If i request the status i would be interested in the difference between the two.
Maybe it could be useful to track if a pin was output or input. But on the other hand it does not matter for the device. So why spent a byte on it?

As for the status structure. Why use 8 bytes? 1 byte would do it?

It's not meant to criticize your work. I would like to learn from your considerations. And maybe mine can add something as well.