Also export the attributes where needed.

So far, the Scorer object only relied on the number of chains in
the target to decide whether to enumerate the full chain mapping
solution space or apply a heuristic. The number of chains in the
model is equally important.

Nonsense results in that context means that the model does not
cover the desired substructure in target.

The default is to rely on residue numbers for ref/mdl residue mapping

A new entity with new residue numbers is constructed for scoring

A few bytes can be squeezed out of secondary structures by using runlength
encoding and transform to integer numbers 0,..7 instead of using the actual
ASCII characters. An even more funky approach has been tried:
introduce 4 bit integers which is sufficient to encode secondary structures.
Two of them can be packed together in a 8 bit integer with some bit
twiddling. File sizes decreased as expected (roughly 3% with respect to
original size). However, after entropy compression, that advantage
diminished, i.e. the simple run length encoding and dumping as 8bit
integers compresses better. Assuming the application of gzip as the common
use case, the 4bit version was dumped. Even though it was pretty cool ;)