chain_mapping/scoring refactoring/bugfixes

- avoids requirement of having views attached to each and every sequence
  that comes out of chain_mapper. Original motivation of this was that
  we're more flexible when it comes to residue number alignments, especially
  when we want to provide custom SEQRES. As an additional treat, we get some
  performance increase. In all-vs-all chain mapping scenarios where all
  structures are already in memory, the creation of all these views requires
  significant resources.
- unit test of Scorer objects
- bugfix in IPS: The reported per-interface ips precision/recall values
  were wrong. The actual per-interface ips score as well as the global
  ips score were not affected.
- bugfix in lDDT: The scorer doesn't use the alignments that come out
  of the chain mapper but create their own. One set of alignments for
  the full model/target and one set of alignments for stereochecked
  model/target. These alignments can be inconsistent if residues in
  the stereochemistry checks are removed at unfortunate locations
  (beginning/end) or simply if a lot of residues are removed. The
  stereochecked alns are not independently computed anymore, but
  start now from the full alignments and just introduce gaps for
  residues that are completely removed. This does not affect scoring
  runs with residue number alignments enabled.