Dirty marking and rendering

Chalk organizes layers in the form of a tree, with a grouplayer at the root:

group1 == image->rootLayer()
    layer2
    layer3
    layer4
    group2
        layer6
        layer7
        adjustmentlayer 1
        layer9
    group3
        adjustmentlayer 2
        layer10
        layer11
        group4
            layer12
            layer13


In this example, group1 is the rootlayer; layer13 is group3 is shown topmost
in the layerbox, with group4 right under that, and layer13 is the "highest"
paint layer in the complete tree.

Compositing

At every group level, a projection layer caches the result of compositing
the layerstack in a projection paint device. The cached projection is then
composited with the layers of that level, etc, until everything is composited
onto the projection or the root layer. The image does _not_ have a projection
anymore.

We composite from layer2 downwards onto the projection of rootLayer, group1.


Dirty marking

In order to do the least possible amount of work (which is very important,
especially with large amounts of layers and adjustment layers), we keep track
of which layer is dirty. Groups without dirty layers are not recomposited; this
dirtiness of course travels upwards, meaning that the rootlayer will always be
dirty.

XXX: Should we keep a structure with dirty rects for every layer, so we
can determine whether the changed rect in a layer is actually in the area
we are recompositing? I don't think so, since we should always try to keep

Adjustment layers also keep a copy of the result of their work; if in group 2,
layer 9 is adjusted, we do not want to composite layer 6 and 7 and filter the
result through adjustmentlayer 1; we want to composite the changes in layer 9
directly onto the cached result of the adjustment layer.