Taken in isolation, this commit may appear heavy-weight. The intent here
is to have a linter run and retain all state information about that run
so we can inspect what happened.

What this commit shows is that the checkers the linter knows about have
the right config, but still multi-job [MJ] runs are incorrect.
    see: assert checker.config.min_similarity_lines == min_similarity_lines

I think the MJ failiures are because SimilarChecker.reduce_map_data()
constructs a new SimilarChecker and then the linter doesn't call
set_option().

The process by which the set_option() calls are made in the normal run
is opaque to me, but perhaps someone else has a better idea(?).

… CLI

Co-authored-by: Pierre Sassoulas <pierre.sassoulas@gmail.com>