Split catalog fixes jax

[talonmyburgh]

Jax has been used to speed up RIMS ~ 50% of what it used to.

cli has been adjusted from ms2dynspec to rims run.

There is now also the option for kronicle publishing runs publish.

Removed killMS dependency and supports using Python3.12 and numpy2.0. Use this DDFacet version.

[talonmyburgh]

I've added some fixes - namely issues #32 #34 #33.

I've also calculated the randomly generated off targets using a CoordMachine (as suggested by @cyriltasse). I then further added some debug so that if targets in the srclist are outside the rad, the message mentions this and how far out they are.

Finally, I've added a --stokes parameter that takes in a str ("I", "IV", "IQUV" etc). This is quite a performance bump (saving ~30-40% of time and RAM) when the dataset was not calibrated for polarisation (currently the state in our tron pipeline). Naturally if parameters U or V are specified, the performance bump goes away as we then need all correlation products. If ignored this parameter defaults to "IQUV".
PS. In making the stokes parameter upgrade, the saved FITS files are also smaller for subsets of IQUV. But this does mean they are non-standard - i.e. ds9 and casa will be angry - however inspect-dynspec is developed to handle these subsets correctly.

[talonmyburgh]

There appear to be issues with the application of the beam in this branch... stalling this PR for now.