witness: freemonoid supplemental Clone-bound impls (row 1a, expected red)

[briansrls]

What this PR is

Witness-only addition for row 1a of the CommutativeSemiring serde-bound E0277 effort (node adhoc-09c47958-545). Adds exactly two new rows to the existing, already-enrolled file dag/test/claim/generic_item_clone_bound_witness_test.dag:

• freemonoid_supplemental_struct_hand_written_impls
• freemonoid_supplemental_enum_hand_written_impls

These exercise the row-1a target shapes directly — a struct and an enum whose only Clone-requiring field is FreeMonoid<T> (repr-grounds to Rc<im::Vector<T>>), matching FreeMonoidUniqueState<T> / ListTailResult<T> in src/v2/std/algebra.dag — through v1_emit_struct_from_capability_table / v1_emit_enum_supplemental_impls end to end, rather than the pre-existing structural/WF fixpoint checks above them in the same file.

No emitter files are touched. Neither forbidden file (src/v1/05_emit_rust.dag, src/v1/trait_derive_emit.dag) is touched. No repair code lands here.

Expected status: BOTH NEW ROWS ARE RED, ON PURPOSE

Both freemonoid_supplemental_struct_hand_written_impls and freemonoid_supplemental_enum_hand_written_impls are EXPECTED RED until the row-1a repair lands. compile.emit succeeds (the emitter does not refuse); the assertions fail because today's emission still puts a blanket T: Clone bound on the item-level header instead of the target shape (bare header + hand-written Debug/PartialEq with a supplemental T: Clone bound scoped to just that impl + #[derive(Serialize, Deserialize)] with #[serde(bound(...))]). This is the discriminating RED the packet cites as evidence the target shape is not yet produced.

Re-verified from a clean instrument (fresh detached worktree at origin/main, zero working-tree diff outside this file, local build) immediately before opening this PR:

• freemonoid_supplemental_struct_hand_written_impls: FAIL (red for cause; compile.emit succeeds)
• freemonoid_supplemental_enum_hand_written_impls: FAIL (red for cause; compile.emit succeeds)
• unbounded_declared_container_negative_control: PASS

Why the pre-existing negative control still passes, and is not superseded

unbounded_declared_container_negative_control (pre-existing row, unmodified here) asserts only item-level header bareness on an unrelated fixture (type FreeMonoidOnly<T> { ancestors: FreeMonoid<T> }) — that the header stays struct FreeMonoidOnly<T>, never struct FreeMonoidOnly<T: Clone>. The row-1a repair this witness targets never touches item-header bareness; it only changes which trait impls are hand-written vs. derived, and adds a #[serde(bound(...))] attribute. So this control's subject is disjoint from the new rows' subject, it keeps passing both before and after the eventual repair, and it needs no retirement receipt.

Enrollment route (confirmed, not assumed)

CI's only witness mechanism is claim_executor --required-floor --source-root dag --source-root src/v2 (gunbc.witness_floor_workflow → .github/workflows/witnesses.yml), which discovers every test fn in every .dag file under the dag and src/v2 roots via run_required_floor (src/v1/stage0/src/cli_run.rs:39590) and folds them through one prepared subject — no plan entry, no selection flag, no manual roster edit required.

This file (dag/test/claim/generic_item_clone_bound_witness_test.dag) is already discovered and already executing its pre-existing rows today. Its module identity (test.claim.generic_item_clone_bound_witness_test) does not match any long_home_prefixes exclusion (test.claim.long., v2.test.long., v2.test.claim.long.), so it is not routed to the excluded long-home lane. The two new test fn declarations are therefore picked up automatically by the existing directory-scan discovery the next time the required-floor fold runs — no additional wiring needed.

This PR was opened in draft (auto-attached by the dashboard); flipping to ready now since the addition is complete and self-contained.

[gunbai-bot]

CI failure on this check is expected and by design, not a defect -- no fix is being pushed.

The witnesses job fails on exactly the two witnesses this PR adds:
required-floor: FAIL test.claim.generic_item_clone_bound_witness.freemonoid_supplemental_struct_hand_written_impls returned Bool(false)
required-floor: FAIL test.claim.generic_item_clone_bound_witness.freemonoid_supplemental_enum_hand_written_impls returned Bool(false)

Both are row-1a "expected red" oracles: they assert the hand-written Debug/PartialEq-impl shape the emitter is supposed to produce for FreeMonoid-containing generic types, and that capability does not exist in the emitter yet. A green result here would mean the fixture stopped testing the real behavior, not that anything improved.

Per smart-ram-730's standing ruling, this PR is deliberately parked as draft and must never merge standalone -- merging it alone would land two knowingly-red witnesses on main with no mechanism to satisfy them, which is exactly what this failing check reflects. It lands atomically via cherry-pick into the row-1a mechanism PR (swift-moth-294's lane) once that capability is implemented, at which point these two witnesses are expected to flip to PASS with zero fixture changes.

No action needed here until that mechanism PR is ready.

[gunbai-bot]

Retiring this PR unmerged. Row 1a was dropped from #8534 (swift-moth-294, by parent ruling) -- it lands as Row 3 only, and E0277/E0369 are already zero on main via #8539's Clone-bound call-forwarding work with no FreeMonoid supplemental-bound mechanism present. The open question that sank row 1a was whether #8539's approach generalizes to the FreeMonoid supplemental case; that question is undecided today and gets settled by execution whenever a real FreeMonoid-field closure next appears, not by landing an untested mechanism now.

That leaves this oracle asserting behavior no mechanism on main implements or is scheduled to implement. Per DESIGN.md's evidence-retirement rule (S3/S4b: a witness whose only subject is behavior the tree doesn't carry is retired with a receipt, never left standing -- otherwise it becomes enrolled-but-can-never-pass, exactly the required-floor CI failure this branch has been showing), the correct disposition is retire, not merge-as-red and not indefinite-draft-park.

What this closes with, for the record: the oracle was verified correct by execution against swift-moth-294's mechanism commits (fce0b6d, 80aea09, 709e72e) before those were dropped from #8534 -- all three rows passed, including the negative control. Four independently-sourced corrections landed on this branch (Copy-on-Rc-field via rustc/std::rc::Rc docs, Debug-prelude/E0404 via doc.rust-lang.org/std/prelude), none copied from emitter output. This branch (witness/freemonoid-supplemental-clone-bound) and its commit history remain as quarry: the discriminating fixture and its derivation trail are here, cheap to resurrect, if/when a real FreeMonoid supplemental-bound site makes the row 1a question decidable again.

Closing without merge.