test(ccb): pin the two frozen byte grammars, so "byte-identical" stops being an unchecked claim

[cryptskii]

The gap

2c-B freezes CloseAuthorizationPreimageV1 and DlvSettleOperationPreimageV1 as byte grammars, then states the authority rule:

The shipping Operation::to_bytes is the current implementation of these grammars and is byte-identical to them. It is evidence of the byte grammar, never the normative authority. A future refactor that changes these bytes changes the protocol and requires an amendment.

Nothing enforced it. Both grammar names appeared in the tree only inside doc comments.

That matters because dlv::close_authorization signs Operation::to_bytes() directly. A reordered field, a widened integer or a dropped length prefix would silently invalidate every close signature already produced, and no test would go red.

What this adds

A class-1 conformance file, per 2c-C2 ruling D: the expected bytes are built by a local encoder written from the amendment's field tables, never by calling the encoder under test. Agreement is therefore evidence rather than a tautology.

The local primitives mirror 2c-B's, which are deliberately the opposite of the canonical encoding's: every 32-byte value is length-prefixed here, because these are foreign grammars carried in a bytes field and not objects in the graph.

Seven tests:

• byte equality against the frozen table for both grammars
• the cleared-signature encoding the close grammar names explicitly as four zero bytes
• a full-length 49,856-byte signature, proving the length prefix does not truncate
• a decode round trip that reproduces the exact bytes
• refusal of a trailing byte

Both grammars were already conformant. These tests record that; they repair nothing.

Mutation controls, executed

Restored from a byte copy each time and verified byte-identical afterwards.

mutation	result
swap owner_devid and owner_genesis in the tag-26 encoder	dlv_settle_to_bytes_is_byte_identical_to_the_frozen_grammar red, plus three more
change the close discriminator from 28 to 27	dlv_close_to_bytes_is_byte_identical_to_the_frozen_grammar red, every settle test still green

The second control also shows the two grammars are scoped apart rather than passing together by accident.

Scope

This pins the grammar. It does not lift the market emission refusal, and it does not wire 2c-B's G1–G4 as production gates. Those need a producer that emits a real preimage, which is 5c-2 Step 2/3. Wiring them now would create a gate no producer can reach.

Verification

• cargo test --locked -p dsm --test frozen_byte_grammars_conformance — 7 passed, 0 failed.
• make lint at repo root on pinned 1.98.0 — passed.
• bash ci/production_safety_checks.sh — all passed, including TLA+.