This is similar to ilddt which is used as target metric in the new
AlphaFold3 paper

and some cleanup...

we differentiate between pair states, i.e. errors related to a specific
pair of model/target ligands, and single ligand states, i.e. errors
That affect one particular ligand. An error state in the latter basically
takes that ligand out of the race for assignment.

This is mainly for debug purposes

That was stupid before

Only initial work done for a base scoring class that takes care
of ligand prep, computing ligand symmetries and calling score computation
functionality from child classes. Child class for SCRMSD is already there
with basic testing.