[kvlonge]

Why

Many planner, plan-output, freshness, lifecycle and selection functions were registered on the native module only so Rust could call them or so Python tests could reach pure Rust logic. Each binding costs build time, stub surface and a module-attribute check, and keeps tests in Python for logic that lives in Rust.

Changes

• 69 functions that only Rust calls lose their #[pyfunction] attributes, registrations, stub entries and module-attribute checks; Rust calls them directly. 11 had no caller at all and are removed.
• Selector parsing, resolution, folder and table-function selection, error messages, SQLBuild model-selector names and unit-test selector splitting are tested in Rust over the sqlbuild-analysis functions that implement them. Every case the removed Python tests asserted is ported.
• The Python tests of the removed bindings, their helpers and an empty test package are removed.
• A dbt plan integration case covers a backslash path: selector through the real binding (folder, translation and exclusion of other folders).

No behaviour, output or diagnostics change.

Verification

• Compiler differential 374/374; planner differential 11/11 against a recording on the base commit.
• plan --json byte-identical on two large synthetic projects.
• Runtime instruction counts at parity (dense compile, warm and cold plan); incremental native build 1.1% fewer instructions.
• Focused Rust tests, about 2.2k focused Python unit, integration and e2e tests, workspace clippy -D warnings, fensu and make check-ci pass.