problem with render() design

[tlambert03]

an issue i'm seeing with the design of Modality with a single render() method: some modalities like SIM really can't render an optical image in full 3D truth space. For sim, while it's technically possible, it would be extremely memory intensive and slow, we're much better off just yielding the 2D "optical images" as needed for digitizing (e.g. at the detector). For example, if we're rendering truth at 25nm axial resolution, we really only need every ~5 planes after convolution. In fact, it's worth looking into whether upscaling in Z is important at all here.