Needs more testing before landing but overall very happy with the direction this is going.
The goal of this series is reducing the overhead of actually packing GPU state. This is orthogonal to dirty tracking, which I also have a branch for. (Make the work cheaper versus doing less of it respectively)