This gets to the heart of the matter. While a bit off topic, we are looking for a method to get the best results. Maybe that is provided by a black box software solution, maybe we can go through the process ourselves with a set of basic tools.
So what is the best sequence? Remove noise first?
If at all possible, that's the best because it will avoid noise propagation by demosaicing into color that pollutes surrounding pixels.
This would be followed by Lateral Chromatic Aberration correction as part of the demosaicing, because the realignment of the RGB signals increasing towards the corners which will help luminance resolution and reduce color pollution when demosaicing.
Then deconvolve, then demosaic,
Maybe, or the other way around, depending on the demosaicing algorithm.
then gamma,
Yes, or stay in linear space for parametric editing and color blending with limited bit depth. When converted to floating point numbers, the back and forth gamma conversions are relatively lossless.
then the ususal assortment of raw tools?
Yes, some will give better results in gamma pre-compensated space while others benefit from linear gamma space (e.g. luminance shifts without color shift), although with floating point numbers it would actually be dictated by whatever requires the simpler calculations in either space.
Cheers,
Bart