fix(tactician): align deterministic rule metadata

[Fartis]

Summary

• align deterministic Judge answers with Tactician completeness, confidence, and evidence-verification metadata
• replace brittle exact-phrase checks with reusable basic-land-type and mana-result concepts
• support both a and an in Oracle patterns such as Each land is an Island...
• add a fictional-card regression using different land types to prove the renderer is Oracle-pattern-driven rather than card-name-driven
• document Sprint 12.3a2 as the closure of the land-type investigation hardening line

Why

The Blood Moon / Urborg / Dryad answer was factually correct and source-grounded, but the public metadata still reported answer_complete: false, judge_verified: false, low confidence, and judge:evidence_incomplete. The contract checker depended on exact sentence fragments and pluralization even though the deterministic renderer had satisfied every requested obligation.

Validation

• python -m pytest -q tests — 320 passed
• focused land-type regressions — 7 passed
• conversational gauntlet — 40/40 scenarios, 58/58 turns
• repository health — 5/5
• release packaging — 1/1
• all focused CI modules passed; executed in chunks because the local sequential runner exceeded the sandbox command limit