"Old" brighter-fatter kernels exist in  /datasets/DC2/repoRun2.2i/CALIB/bfkernels/​.  Unfortunately, these are in an intermediate format that as far as I can tell, were only ever used in this one place.  Reading the kernels is no longer possible with the current stack, as the pickle storage for the kernel attempts to map them to the current style.  Hacking with ip_isr and cp_pipe at w.2020.15 (a date I'm reasonably certain had the old code) allows me to access the kernel, and can demonstrate that two randomly selected kernels have identical values.  This indicates that there is only one set of kernel values, so constructing new kernels should not be difficult.  I've attached the kernel value to this ticket.

Plot revealing the lack of correction from the 2.2i/runs/test-med-1/w_2021_16/DM-29899: gen3 run:

Based on the comments on DM-26092, this should be working now.

I've split this review into two parts, with `ip_isr` for Andrés Alejandro Plazas Malagón, and the remaining PRs for Dan Taranu.  

Looks good. I tested locally with ci_imsim and had one jenkins build pass (the macos timeout and failures in other runs are not related to this ticket).

Looks good. My main question was if the changes here were only related to the logic for AMP and DETECTOR levels and how that is related to fixing the BFKs for DC2. Chris confirmed that, in fact, these are only AMP/DETECTOR differences to match what gen3 is expecting (a single kernel per detector).