try_updates: when an RA,Dec step is huge to the point of ridiculous, drop further steps

minor: int halfsize

galaxy models: add force_sourceOut arg, to force rendering of models when the source is outside the modelMask, and use when computing derivatives.

warn but proceed with huge galaxy models

Merge pull request #143 from dstndstn/main-gpu-cache

fix cython compile error

remove commented-out blocks

up-front assess if clipping will occur and call non-cached version

pre-compute range of dx,dy values that will work

ugly: explicitly monitor for clipping w/ cached data and fall back to no-cache

logprob -> dlogprob typo

fix offset vs cached image_params; compute delta-chi-squareds directly

Merge branch 'main' into main-gpu-cache

fix the logic for ramping between FFT and MoG components

investigate PSF fitting taking a large number of steps