v1 emitter: extend Clone-bound WF propagation to FN/IMPL emission

[gunbai-bot]

Worker attestation

• Title describes the change (not the session id or branch).
• PR body summarises what and why.
• Tests run: named below.
• No commits on this branch are surprises.
• No secrets / credentials / large binaries staged.

Summary

Follow-up to #7691: extends v1's Clone-bound well-formedness propagation from struct/enum items to fn declarations and fixes a second, related bug in impl-header emission for enum shared accessors, both discovered live in deep-heron's honest regen of dag/std/occurrence_binding.dag.

Two distinct bugs, one root mechanism:

1. E0277 (bare fn generics). A fn naming a Clone-WF-bounded declared type in a value-param or return-type position (e.g. occurrence_binding_from_candidates<N>(occurrence: BindingOccurrence<N>, ...) -> OccurrenceBindingResult<N>) needs N: Clone for the same reason a struct field does, but the fn path never checked it. Root cause was a raw-root/resolved-children inconsistency: the fn-path WF check uses the raw (un-typechecked) top-level param type_expr (needed because the typechecked node loses ident_span for fn params specifically), but three downstream recursive helpers resolved nested children via child_type_node, which loses ident_span again — a mismatch specific to the fn path (the struct/enum path is unaffected because it's raw-or-resolved consistently top-to-bottom). Fixed with one shared fallback helper, v1_wf_child_type_node: try the resolved node, fall back to the raw child only when the resolved node's name is empty. This is a single-authority fix, not a forked predicate, but it is not confined to the fn path: the three call sites it replaces (v1_type_expr_clone_impl_needs_param, v1_declared_arg_positions_need_clone_param, v1_type_expr_wf_needs_clone_param) are shared helpers also reached from the item path via v1_item_param_wf_needs_clone, so struct/enum recursion now runs through the same fallback too. What is unchanged is the item path's output, not its code path — the fallback is a no-op there because struct/enum resolved children always carry a non-empty authored name. That is an invariant, not a re-derivation: the evidence for it is the nine pre-existing item witnesses still passing unchanged plus deep-heron's 58-to-0 production regen (below), not a proof that the invariant must hold for every shape.

2. E0229 (impl-header bound repetition), found while writing the witness for (1). emit_enum_shared_accessors rendered impl<N: Clone> Foo<N: Clone> { ... } — the Clone-bounded type_params string (with bounds, correct for the impl<...> declaration position) was reused verbatim for the type-application position (Foo<...>), where a bound on a type argument is invalid Rust. Added emit_bare_type_params to render the name-only form for the type-application position; the declaration position keeps the bounded string.

Dual-representation audit (per review request): neither fix re-derives Clone-bound facts independently. The fn-grain check reuses v1_type_expr_wf_needs_clone_param verbatim against the same emit_info.clone_bounded_type_params — the single item-level fixpoint result computed once per compile by v1_clone_bound_fixpoint_loop (landed in #7691). The fn path runs no fixpoint and holds no bounds map of its own; it only asks the existing authority a WF question at a new call site (fn params/return, the "third site" alongside struct fields and enum-coproduct fields). The impl-header fix touches only string rendering (which positions get the bounds string vs. the bare-names string) — it does not touch bound computation at all.

Availability note: Codex is out until Aug 8; this PR is unreviewed by that provider in the interim.

Test plan

• cargo run --manifest-path src/v1/stage0/Cargo.toml --bin regen_stage0 — clean regen, 113 files, no diff beyond the intended .dag source changes reflected.
• cargo run --manifest-path src/v1/stage0/Cargo.toml --bin claim_batch -- --source-root dag --source-root src/v2 --entry dag/test/claim/generic_item_clone_bound_witness_test.dag --functions fn_wf_bound_propagation_positive,fn_wf_no_trigger_negative_control,impl_accessor_inherits_item_bound_once,containment_path_clone_bound_positive,unused_generic_param_negative_control,both_triggers_render_single_bound,declared_generic_type_propagation_positive,coproduct_well_formedness_propagation_positive,coproduct_bare_payload_negative_control,unbounded_declared_generic_negative_control,recursive_declared_type_saturation_positive,unbounded_declared_container_negative_control — all 12 witnesses (3 new + 9 pre-existing) PASS, exit 0. The 3 new witnesses are RED-on-pre-fix/GREEN-on-fix, executed both directions: confirmed FAIL before the fix (fn path unbounded / impl header double-bound) and PASS after.
• Direct specimen check: gunbc compile --source-root dag --entry dag/std/occurrence_binding.dag --target rust now emits pub fn occurrence_binding_from_candidates<N: Clone>(...) and impl<N: Clone> OccurrenceBindingFoldState<N> { (previously <N> bare and impl<N: Clone> OccurrenceBindingFoldState<N: Clone> respectively).
• Acceptance oracle, run by deep-heron-234 against this PR's exact head a283ebdc3eda062096e4606e773cfd78dabd772c (stale-binary discipline: fresh regen_stage0 binary rebuilt from that head, never a reused committed binary): worktree at a283ebdc3ed → rebuilt regen_stage0 from that tree → emitted occurrence_binding + candidates/resolve/parser_walk/witness from deep-heron's B1 authored sources with that binary → installed the emit into the real v1-compiler stage0 crate → cargo check -p v1-compiler. Prior Propagate item Clone bounds as a fixpoint over the declared-type graph #7691-residual class was 58 errors (E0229 double-bound impl + E0277 bare fn generics); after this PR's emit, 0 errors, clean Finished, with the specimens confirmed correct: impl<N: Clone> OccurrenceBindingFoldState<N> (no double bound) and occurrence_binding_from_candidates<N: Clone>(...). No residual Clone-emit error class found; no third emitter PR needed for this gap.

Owed follow-up (not a blocker for this PR): coverage of v1_wf_child_type_node is asymmetric. The fires-branch (resolved child empty, raw child used) is discriminatingly witnessed by w_fn_wf_bound_propagates_to_bare_generic. The no-op branch (resolved child non-empty, taken on the item path) is covered only indirectly — the nine pre-existing item witnesses continue to pass, but nothing in the suite would necessarily red if the fallback wrongly fired on an item, since taking the raw child yields the right answer for many shapes anyway. A witness that discriminates the two arms directly belongs in a follow-up PR, where it costs nothing to add; extracting it here would restale both approvals and cost another CI cycle for something the production oracle already exercises in practice.