Updated Transform so it accepts a greyscale input per parameter. For each pixel, the parameter is multiplied by 2*v-1 (v being the value of the input's pixel) which makes it possible to affect the strength of each transform using an input image. New effects such as whirl and color guided offsets are now possible.
Added Decompose node (does the opposite of combine).
Used another color for inputs/outputs that expect/generate a greyscale image.
* updated transform node (scale split into scale_x and scale_y, and added a repeat parameter)
* added blend modes (burn, dodge, lighten and darken)
* added a (very bad) blur node, based on generic convolution code
* rewrote normal_map node using convolution code
