[jasperblues]

Closes #427. Port of liberation-data/drivine4j#612 (drivine4j PR #623), as liberation-data/drivine#638 ports it. Not to merge until train 2026.11.1 is cut, and not before drivine#638.

On Postgres, SQLite and Oracle without Oracle Text, a prefix term (kiw*) was weighed by BM25, while Neo4j's Lucene scores a prefix query as a constant. bm25_ranked (drivine/backend/sql/bm25.py) now adds PREFIX, k1 + 1 = 2.2, to each node that holds a prefix term (is_prefix). That is Lucene's constant of 1 on bm25_score's scale. Every other term scores as before.

Goldens: unchanged, since none of them pins a prefix query.

Tests

• tests/integration/test_bm25.py, and its sync copy, on Neo4j, Postgres, Postgres indexed, SQLite and Oracle:
  • kiw* scores every hit 1.0;
  • kiw* bird ranks the node with both first;
  • a Fruit label with #612's texts: kiw* and kiwi* give every hit 1.0, and kiw* melon and +kiw* melon give the formula's figure, within 0.01 of Neo4j's measured 0.5409. On Neo4j, with lengths in tokens, it's within 1e-4.
• tests/unit/test_bm25.py: the score SQL for kiw* melon.

Verification

• Two fresh-reviewer passes. Pass 1 added the unit case and the docs notes (docs/02-connecting.md, docs/07-schema.md); pass 2 came back clean.
• Unit tests: 2,471 passed. ruff and pyright are clean, and the sync code is current.
• Integration, on the test box with Oracle: the BM25 tests, their sync copy, and the search tests. 207 passed, and agreement with Neo4j is unchanged at 17/21.