According to your suggestion I just debayered my calibrated data as well with the Bilinear as with the SuperPixel interpolation method. The result is: It makes no difference in respect to the colored background artifact in the resulting integration.
I never tried bayer drizzle, so I have no experience how to apply it to my data.
I don't quite understand this statement. I never used an output pedestal and didn't experience any drawback so far. When my light frames are calibrated with my 150 s MasterDark and the MasterFlat, there is no severe clipping to zero. I have checked this again: with the set of 95 calibrated light frames the average number of pixels that are clipped to zero is 29 pixels. Given that the sensor has 11.7 million pixels, this is negligible. I can only imagine that you are referring to the wrong MasterDark (300 s) that I uploaded by mistake in the first zip archive. The 300 s MasterDark indeed could have produced some clipping when applied to the 150 s light frames.
Anyway, I did as you suggested and performed the workflow with an output pedestal of 800 DN applied. Again there was no visual difference in the integration result, and their histograms (workflow with / without pedestal) are barely discernable. So this is also not the cause of the colored background artifact.
Definitely not the wrong dark. I created a new dark from the 150s frames, then used that new master to calibrate each of the same 150s frames. Without a pedestal, they did not calibrate properly. I would try using a pedestal when calibrating your lights and see if it changes anything.