memory consumption due to static ncg jit

I noticed a much higher memory consumption which was introduced here (b7404bf0)

You can try it yourself by running 1_tomography.py on this commit and the commit before (e3ead6d8)

As far as I understand the interface of optimize_kl, the user can change back to _newton_cg,

However, is it possible to have less memory consumption in the _static_optimise_kl (probably introduced by jit)?

@pfrank @gedenhof