Experiment with `atx` for FINK

[tallamjr]

It was noticed when reproducing plots, that atx seemed to be faster for inference. This prompted a thought that it might be useful to see if atx achieves similar performance to t2 when trained for FINK (i.e. only g and r bands with no redshift information -- currently t2's LogLoss ~ 0.96 in this setting)

If atx performance is comparable, then copy of https://github.com/tallamjr/fink-science/blob/issues/pending/t2mod/fink_science/t2/t2-MWE.ipynb should be done, but for atx to weight up which model would be best for deployment.

[jasonmcewen]

Hmm, I'm not sure this is worth doing. Let's discuss further in our meeting next week...