Pathfinder doesn't respect `vectorize` argument in idata postprocessing

[jessegrabowski]

I have a model with very large constants that blow up my machine if they get broadcast. Pathfinder estimation works great in PyMC mode, until we get to here, at which point it hangs and the kernel dies. This should sample sequentially if the vectorize = False kwarg was passed.

[ricardoV94]

Wonder if FAST_COMPILE is the culprit, not giving it chance to memory optimize

[jessegrabowski]

Also not great