Astrometry offset

[tboch]

Astrometry accuracy is degraded in v3.4.5 wrt 3.2.0

Here are 2 examples illustrating the issue.

1. Cassiopea constellation drawing (example https://aladin.cds.unistra.fr/AladinLite/doc/API/examples/polyline/ )

With v3.4.5:

With v3.2.0:

2. Zooming on object * bet UMi ( 14 50 42.326 +74 09 19.81 )

Centering the current view on * bet UMi, with AITOFF projection (but the issue is present in all projections, as far as I can tell).

With a 10° FoV , the star is properly centered:

With a 32° FoV, the shift is obvious:

@bmatthieu3 : could this be linked with some of the astrometric computations having been offloaded to the GPU, when previously

done on the CPU?

[bmatthieu3]

The accuracy is enhanced since #180 but there is still some deformation present near the healpix poles. We could subdivide the cells but if we do that we must subdivide all the cells in the fov in the same way otherwise we would see frontiers between cells.