refactor: host migration, cursor and fixture planning in Rust

[kvlonge]

Why

Continue Rust-hosted planning while preserving migration safety, SQL artifacts, adapter overrides and persisted identities.

Changes

• Native migration compatibility, discovery, model/column decisions and origin/rendering orchestration.
• Native reference/source and model/function/hook resolution, cursor bounds, overrides and safety policies.
• Native fixture alignment, formatting, schema/type planning and scenario graph/relation planning.
• Preserve Python adapters, executor and boundary models; delete replaced implementations. Movable Python decreases from 39,461 to 33,506 lines.
• Keep fixture type metadata shared: extract only selected-chain inputs rather than copying the whole project for each SQL test.

Verification

• 11,412 unit/integration tests passed; all-extras static CI checks passed.
• Compiler differential: 371 projects, required discovery/render coverage and unchanged goldens.
• Planner differential: 11 projects across 18 stateful steps matched the existing baseline.
• Cold/edited compile output digests and exit codes match the reference on synthetic and private compile-only projects.
• Independent review identified project-wide fixture metadata copying; standalone fix passed 19 focused tests, updated static checks and bounded follow-up review.
• Deterministic regression: growing unrelated metadata from 10 to 10,000 relations leaves conversion at one relation/two columns, preserving shared base identity and nonmutation.
• Worker comparisons cover migration decisions/hashes, adapter callbacks, cursor timezone/boundary cases, fixture SQL and scenario plans.