[kvlonge]

Why

Phase 1 of the Rust-hosted migration moves compiler logic into Rust and deletes the Python it replaces. This milestone covers the compile-input stage: preparing each model before the native model loop, plus the public declaration indexes, required column reads, and default layering that sit around it.

Changes

• The native model loop now does @@ interpolation and declaration scans itself, builds the known resource names from the discovered inputs, and indexes SQL hook definitions.
• Public enum, constant, and reusable model schema indexes (extends resolution, cycle and override errors) are built in Rust.
• Required column reads used by fixture completion run on the polyglot crate directly. This replaces a Python walk over the polyglot wheel and keeps the wheel's accessor semantics.
• Project default layering and the description resolution used by lint run in Rust.
• Deleted the unreachable legacy schema.yml model-entry attachment checks (discovery always rejects models: content), plus dead compiler Python.
• New goldens: three differential failure cases cover model schema extends errors. Each golden is added for a new case; no existing goldens change.

Movable Python: 143,789 → 141,910 lines.

Verification

• Compiler differential: 371 projects pass, and the goldens only add the three new failure cases.
• Python vs native required column reads agree on 4,000 model/dialect pairs and 204 edge-case queries.
• Compile output is identical to main on a 1k-model project and a real project, both cold and after an edit.
• make test (11,572 passed), make check-ci, and the public-repository scan all pass.