Add nonrigid_lowrank transformation

[gadomski]

Andriy's reference implementation came with a nonrigid_lowrank transformation, which used some tricks to speed up the very slow nonrigid registration. I think we could use https://github.com/m-reuter/arpackpp to do the eigenvalue calculations required, which is the only real hurdle (for now).