afterburners register/launcher framework

[julienguy]

We could add to the pipeline the framework to run afterburners

• with an abstract (template) class with functions run() read() write() ...
• with the same inputs (suggestion = a coadd + a zbest file)
• similar output (suggestion = an astropy table with at least the TARGETID as a column, the rest being free, maybe with restriction on types, int or float, no arrays, no character string because it can be a mess).
• tables of all the registered afterburners would be automatically merged

This would be similar to what we developed for nightwatch QA, see https://github.com/desihub/nightwatch/blob/master/py/nightwatch/qa/base.py

We would write in this framework wrapper classes for quasarnet , the Mg line fitter for QSOs, the [OII] doublet line fit for ELGs ...

[sbailey]

Another thing to consider is defining parallelism, memory, and runtime needs. We had a hard enough time sorting that out with desispec.pipeline.tasks.base.BaseTask with pipeline pieces where we had full control; it could get messy with end-user contributed afterburners.