Follow-up to the PerformanceMetric dedup: make the performance-metrics taxonomy
use proper unece: linked-data instead of imported untp: literal qualifiers.

- Re-add unece:aggregationMethod / improvementDirection (object properties ->
  the aggregation-type / improvement-indicator schemes) and allowedUnit
  (-> unit-of-measure). Complete the two hand-authored schemes with the values
  the taxonomy actually uses (aggregation-type +average/count/maximum/minimum;
  improvement-indicator +context-dependent).
- build-untp-taxonomies.py converts each concept's untp:aggregationMethod /
  improvementDirection / allowedUnit into the unece: property with a concept URI
  (allowedUnit falls back to a literal for codes outside our Rec 20 subset, e.g.
  MK / P1 = percent, which is filtered from the generated units).
- Fix two latent generator bugs surfaced by the re-run: it emitted the source
  production namespace (now rebased to the authoring BASE) and dropped the
  codeListScope tag (now emitted). Both taxonomies are domain-scoped.
- aggregation-type and improvement-indicator are bound again, so revert their
  ledger rows B -> A.

validate passes; diagrams/docs/site regenerated.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>