In principle we have three types of structures in the scorer (terrible I know).
1) the actual strucures that are only molcked 2) the structures as they are used
in the chain mapper (complete backbone) 3) same as 1) but stereochecked.
The alignments required at this position are of category 3 but they were 2.

This code is considered deprecated as of release 2.4.0. The new lDDT
implementation has already been integrated into
qsscoring.py/compare-structures. I prefer to leave this completely legacy.

Only the fast residue number alignment has the requirement of no insertion
codes and strictly increasing residue numbers.

That was 1) strictly increasing (no idea why it was there in a first place)
2) no ins codes (The residue mapper is a dict which now has actual residue
numbers as key - can handle insertion codes)

RepResult was fixed to not throw errors any more

The code would previously only consider the last chain mapping in
consideration. As a result we find better scores in tests.

fnat, lrmsd, irmsd etc. are not strictly DockQ related, they just
serve as input for DockQ

Raise a ValueError upon initialization.

Fixes the SDF reader and documents the convention.

Reason is, it's unclear if they are useful anywhere else