Consistency in the relation to length or size in arguments of generators

[amirDahari1]

It would be more convenient if all generators features would be either relative to the lengthscale of the image (shape.mean()) or to the size of the image (np.prod(shape)). For example, the blobiness argument in blobs is relative to shape.mean() with the sigma of the gaussian blur defined by np.mean(shape) / (40 * blobiness), while ncells in voronoi_edges is relative to the size of the image (in order to get the same size of cells in different images, we need to multiply the number of cells by the size of the image and not by the lengthscale).

I think it would make more sense to have all arguments relative to the lengthscale, so that the comparison between 2D and 3D would be easier and straightforward.