Analysis release 2.0.1

Fixed 🐞

• Two domain maps can now be merged. ccsh merge refused any pair of inputs that both carried domain words
  (Opaque lens 'domain' has conflicting payloads), so domain-parsing a monorepo one subproject at a time and
  merging the results was impossible. The lenses are now unioned by node id, and a word both inputs describe keeps
  the higher frequency and the higher tfidf independently — the same "highest value wins" rule ccsh merge
  already applies to metrics. A word only one input scored keeps its score. --large merging no longer refuses
  either: it re-keys the domain lens onto the wrapped paths instead of failing, so the words follow the files
  into their prefix folder.

Chore 👨‍💻 👩‍💻

• Cleared the SonarCloud code smells the analysis project had accumulated: CSVRow reads its columns through the
  indexed accessor, InputHelper.isInputValidFolderOrAnyFile reads as guard clauses, UnifiedParser takes
  TreeSitterExcavationSite from the version catalog instead of a hardcoded coordinate, and seven stale TODO
  comments were removed. No behaviour changed.

• The domain lens is now a typed part of the model alongside metrics and dependency instead of an opaque JSON payload, so
  filters can read and rewrite it rather than pass it through blindly. The emitted .cc.json is unchanged, except that a run
  that found no words at all now writes the reserved empty "domain": {} instead of "domain": {"nodes": {}} — the two always
  meant the same thing, and only the latter used to count as carrying data.