Support more dates and more targets for comparison with jplephem

[rhannequin]

This greatly increases support for accuracy validation by comparing with the Python library jplephem.

Now, we have the following parameters:

• all days from 2000-01-01 to 2050-01-01
• kernels de405, de421, de430t, de440s
• targets from 0 to 10

This represents 730,520 combinations for which we compare the position (x, y, z) and velocity (vx, vy, vz) with jplephem.

The Rake task has been redefined to take parameters so that we don't test everything at once, but only 18,263 lines by combinations of parameters. This was made so that validating accuracy could be run automatically on CI in a short amount of time (less than 3 minutes). Everything is declared in validate_accuracy.yml.

The great news is that 4,383,120 values (3 for position, 3 for velocity, every day from 2000 to 2050, 10 targets and 4 kernels) match with jplephem within 2 centimeters or less.

To keep them light-weight and performant, kernels have been reduced to excerpts using jplephem command-line tool:

python -m jplephem excerpt 2000/01/01 2050/12/31 de405.bsp de405_excerpt.bsp

By the way, this is a function that I would like to implement into Ephem as well, one day.

This reduces their weight by:

• de405 from 65.4 MB to 5.6 MB
• de421 from 16.8 MB to 5.6 MB
• de430t from 133.6 MB to 6.3 MB
• de440s from 32.7 MB to 5.6 MB

The tested dates have been limited from 2000 to 2050 to keep execution time low and CSV files "short". We are entering 2025 to this is testing exactly 25 years ago and from now, which seems like a good compromise, but we could consider expanding coverage at some point.