[kvlonge]

Why

Native orchestration still consumed Python executor objects after execution, and persisted cost reporting unnecessarily started Python. Obsolete migration exports also kept unused production paths alive.

Changes

• Use typed Rust executor results for build/dbt completion, progress, JSON output, freshness, runtime SQL and integration events while retaining Python-owned interfaces.
• Share Rust cost storage, selection, exact-decimal sorting and rendering; keep warehouse refresh at the existing adapter edge and never retry output writes through Python.
• Remove fourteen obsolete exports and exclusively used support; preserve historical parity fixtures and gate legacy analysis batch helpers to tests.
• Lower crossing budgets without expanding Python stays, warehouse state or public API ownership.

DuckShop additions are not applicable: the regressions concern CLI/result serialization and display, not project shapes, data or warehouse lifecycle behavior.

Verification

• make test: 12,185 passed before the final dead-binding cleanup.
• Rebuilt extension/launcher targeted regressions: 227 passed, including seventeen real-launcher cost cases and DuckDB/compiler integration paths.
• make check-ci, focused Rust analysis/host/cost tests, hygiene and whitespace checks passed.
• All sixteen original JSON baseline documents are byte-identical; independent review and scoped follow-up found no remaining blockers.
• Windows, warehouse and full E2E coverage remain CI gates. No live compiler-analysis semantics changed.