Handle relative astrometry properly for massive companions

[sefffal]

Right now we assume that the planets have no impact on the star's position.
But astrometry of planets is measured relative to the star, and the planets to affect the stars position. Massive planets can tug the star a few mas away, which is more than the uncertainty on some astrometry measurements.

We should compute the impact each planet has on the star's position and adjust the measured astrometry appropriately before computing the likelihood,

[sefffal]

This is now complete for relative astrometry. Still TODO for sampling from images.