feat(007): Hyperlexical card rename for seed-morph78

[scrimshawlife-ctrl]

Summary

Step 1 of the post-name_gate sequence: the card rename. Follows amendment A6 (#103). No Hub, no T13, no new climb.

• Hub card (hf-package/config.json + README.md): model_id → hyperlex-structure-149m (locked C31), name_gate: true, e2_pass: true, last_trainable: 8, pin: seed-morph78. Still not uploaded.
• New scripts/shadow/hyperlexical/name_gate.py (torch-free): APPROVED_PINS = {"seed-morph78"}, resolved from the real model directory, so a moved BEST symlink can't name another checkpoint.
• eval_unbind: name_gate is true only for a trunk-forward eval of an approved pin. Stub, digest, abort, and any other checkpoint stay false.
• Eval schema: name_gate goes from const false to boolean.

Deliberately unchanged

• contracts/training/v1.schema.json and the save_pretrained HF dump writer: every training run produces a new, unnamed checkpoint, so name_gate: false there is still correct.
• Local train-out paths (~/.hyperlex/models/hyperlex-encoder-modernbert-base-seed-*), MODEL_ID_SEED, HYPERLEX_*, package names (docs/NAMING.md: no hard-cut).
• Existing receipts; amendments.md A6 text (historical scope).

Validation

• New tests/shadow/test_hyperlexical_name_gate.py: only morph78 approved; symlink resolution; digest eval of the pin dir stays unnamed; schema boolean; card config matches pin; module torch-free.
• pytest -q: 738 passed, 6 skipped. Shadow suite 373 passed.
• hyperlex.py check / doctor / smoke: PASS. mkdocs build --strict after CI sync: PASS. ROADMAP mirror matches.
• NOT_COMPUTABLE here: a trunk-forward eval on Spark producing name_gate: true. It will be exercised in the inference-path PR (step 2).