Factorize algorithms from ntuple writer

This started out as a simple attempt to run the main algorithms without necessarily writing an ntuple, and sort of blew up.

Bonus things:

• I wanted better unit tests, so I added a script to run them locally (without access to cvmfs). I also created a public repository that stores the input files on git lfs. This should probably be moved to the same group as this repository, and be given a better readme to explain how to update it.
• I added something to cache the metadata as a json file locally. It's off by default because it could probably confuse people, but it cuts the time to process 10 events from 30 seconds to 10 seconds, so it's very nice for debugging.
  • Note that I also had to disable muons for this to work: when I used the caching all the speed up was lost when the muon configuration hung for 20 seconds. I think they are accessing the same configuration somehow but I can't figure out where.
• I fixed a bug in the help formatting
• The CMakeLists.txt file had a lot of wildcarding. This isn't really recommended last time I checked: the problem is that if you add a new file CMake won't know to rerun.
• I added __init__.py files to the python modules. I don't really know why this was working before but this seems more standard.
• The CI now uses needs everywhere, so the execution is a proper DAG.
• Deleted an unused _load.cxx file, these aren't needed in release 22.