feat(text): real article-level text via normas.leg.br JSON-LD tree

v0.2.0 shipped identification (br_get_norma) but flagged full article text
as an unconfirmed gap - no mechanical rule mapped a URN Lex to a Planalto
full-text URL. normas.leg.br/api/public/normas closes it from a different
angle: a schema.org Legislation JSON-LD tree on the same domain as the
citation page, with real inline text per article/paragraph, addressed by
the same URN Lex - no scraping, no invented URL rule.

Adds text_client.py (HTTP) and norma_text.py (tree walker, clean-room
against the live response) plus two tools: br_get_norma_index (addressable
structure) and br_get_norma_texto (one article's real text). Verified
against the Codigo Civil (2511 nodes, Art. 1 text matches word for word).

Also adds CONSTITUTION.md (was missing - the non-EU connector line isn't
covered by factory_health.py yet, a separate gap).

DISCOVERY.md/README.md/SOURCES.md restructured so current status is
unambiguous, with version history kept separately for the record.