demo tutorials for building intuition of forward model

[geoffwoollard]

I think it would be good, in a demo notebook, to show when aspects of the forward model break down and artefacts to look out for.

• if translation is done by Fourier phase shifts, when the high frequency is numerically unstable (large translations)
• the difference including hydrogens makes (pdb based forward models)
• aliasing effects with high defocus CTFs (relative to small pixel size)
• PDB coordiante centering and voxel / image locations - going back between atom coordinates and map. how to fit them and where the origin is
• run time and memory bottlenecks in PDB coordinate vs voxel forward models. quantify computational savings.
• check if gradient information is propagated in parts of forward model (some discrete random sampling methods, e.g. Poisson sampling, or sparse operations may not have gradient methods implemented)