More than bilinear centering of the model for PSF 2d modeling

This ticket is from the series ``Improvements to the PSF 2d modeling, April 2019 brainstorming''.

The centering of the model is done very crudely at the moment. The model at the moment creates many images separated by 15 microns (which is equal to 1 pixel) and determines which one is best. Then I search for the superpositions of two nearby images which gives best results (lowest chi**2). In conversations with Jim and Robert we have concluded that this is very similar to doing crude bilinear interpolation. These can and should be done better with some proper interpolation techniques (sinc, Lanczos and company).