remove trunc option from moffat

[beckermr]

It turns out this option is very problematic. For example, we can't take derivatives with respect the scale_radius of the moffat profile if trunc is non-zero. The issue is that convolutions track whether or not profiles have hard edges and this causes it to attempt to branch on a traced value.

[beckermr]

Another option is to declare it as static metadata when we construct the pytrees. Then we could support most of what we have now with the caveat that the trunc values have to be fixed at compile time for jit.

[beckermr]

Galsim only requires trunc for beta <= 1.1.

@rmjarvis Do you have a sense if the galsim Moffat trunc option is useful for beta > 1.1? Maybe you have some history of the trunc option that can help provide context?

[rmjarvis]

I believe the trunc is mostly of historical interest. Very old simulations with pre-GalSim tech often truncated profiles for memory reasons. We included it to enable comparisons with those sims. I think it would be fine if GalSim-Jax didn't bother including this feature. But I'm disinclined to remove it from the main GalSim, since I don't know how many people have used it in the wild.

[beckermr]

xref: #196