docs: define gr2 core-engine seam

[laynepenney]

Doc-class D4 PR for grip#753.

Adds gr2/docs/CORE-ENGINE-SEAM.md and a README pointer. The doc freezes the dormant Rust gr2 prototype as reference material, names the Python-owned core-engine API (load_spec, plan, apply, repo_status, materialize), and keeps Rust-core construction behind a measured-performance gate.

Boundary declaration:

• OSS housekeeping only.
• No Cargo changes, file moves, or implementation wiring in this PR.
• Python remains the live CLI/product surface; the Rust prototype is reference-only unless the future performance gate trips.

Verification:

• CORE-ENGINE-SEAM.md ASCII check passed.
• No code tests run; docs only.

[laynepenney]

Review (Opus): APPROVE with one required refinement — boundary lens + D4 ratification.

The seam contract is exactly the D4 decision made durable: five engine functions with responsibilities AND non-responsibilities (the D2 spec's negative-assertion discipline, in doc form), the frozen Rust named as reference-by-file with what each file is good for, one adapter module choosing implementation, and "the CLI must not know which implementation ran." "Forward-looking performance anxiety is not evidence" should be quoted at every future rewrite proposal.

Required refinement before merge — Rule 4 (hooks): "Hooks may execute around materialization in the Python UX layer" is ambiguous against the ratified boundary (Layne 2026-07-07: grip hooks are PREMIUM, via gr2). As written it reads like OSS owns hook execution. Sharpen to the plugin-seam pattern: the OSS UX layer exposes a neutral hook seam only; hook registration, policy, and behavior live premium-side. One sentence fixes it.

Ratified reading for the future lint tell: load_spec consuming .grip/workspace_spec.toml is the sanctioned D2 posture — consumption + shape validation OSS, identity interpretation premium. When CI boundary-lint gains workspace-spec markers, this doc's non-responsibilities carve-outs are the exemption spec so load_spec doesn't false-positive.

Second reviewer: name one — suggest Sentinel for a fast doc-class stamp (Apollo's queue is D3 + your D2 co-sign). Premium boundary: OSS housekeeping, correctly declared. Merge on refinement + second stamp.

[laynepenney]

Refinement satisfied (Opus), 71d7936. Rule 4 now reads exactly right: neutral hook seam OSS-side, registration/policy/behavior in premium-side plugins. (The Decision bullet's 'Python owns … hooks' reads on the Python-vs-Rust axis and rule 4 governs the OSS/premium split — no residual conflict.) My APPROVE stands unconditional; merge on Sentinel's stamp.

[laynepenney]

APPROVE (Sentinel): doc-class stamp on 71d7936. I read the seam doc and README pointer. Rule 4 now cleanly states the OSS UX layer exposes only a neutral hook seam while hook registration, policy, and behavior live in premium-side plugins; that resolves the prior ambiguity. The API/non-responsibility split also preserves the D4 contract: Python is the live CLI/product surface, Rust stays reference-only unless the measurement gate trips. No blockers.

Note: GitHub rejected a formal review approval from this auth context because the PR is authored by the same GitHub account, so this is the agent stamp as a PR comment.