Derive Ord bound for fn-generics reaching set_union: audience_subset/audience_join

[gunbai-bot]

Summary

• v1's fn-generic bound-inference mechanism (v1.trait_derive_emit / v1.compiler.trait_bound_witness) only ever derived Clone bounds for generic function parameters. A generic reaching the set_union builtin — which lowers to v1_rt::rc_set_union::<T: Ord + Clone> — emitted without its own P: Ord, producing E0277 at the call site.
• audience_subset<P> and audience_join<P> in dag/std/authorization_profile.dag are the two real specimens (E0277 root A residual: dashboard node://adhoc-a9f61ade-340).
• Adds a narrowly-scoped third trigger alongside the existing Clone-bound machinery: a pure decision core (v1_set_union_ord_bound_param_names, trait_bound_witness.dag), a Node-consuming body walker locating set_union call sites and resolving their element-type argument names (05_emit_rust.dag, mirroring ownership.dag's collect_callable_refs, plus an ExprBinOp arm that precedent lacks — needed because audience_subset's set_union call sits inside a == comparison), and a sibling renderer (v1_emit_type_params_with_clone_and_ord_bounds, trait_derive_emit.dag) wired into emit_fn_def alongside the existing Clone-only renderer.
• Scoped to the set_union builtin's known lowering only — not a general trait-bound inference system, and does not touch the separate Root A Ord-5 population on EnumeratedAudience's own struct derive (different root, out of scope here).

Verification

Since the built gunbc binary compiles v1's own .dag sources ahead-of-time (it does not interpret src/v1/*.dag at runtime), verifying this required the full self-host regen + rebuild loop:

1. claim_executor --required-regen against these .dag edits (writes a candidate stage0 Rust mirror reflecting the change — reports drift, as expected, since the committed mirror predates this fix).
2. Swapped the candidate mirror in for src/v1/stage0/src, rebuilt gunbc from it.
3. Ran the candidate gunbc compile --target rust on dag/std/authorization_profile.dag.

Result: audience_subset<P: Ord + Clone> and audience_join<P: Ord + Clone> now emit correctly. All other Clone-only generic signatures reachable in the same import closure (ordering_fold, effect_access_request, list_length, disclosure_access_request, etc.) are unchanged — no regression on the existing 25-site Clone population.

The committed stage0 mirror itself is not touched by this PR — it is expected to converge via the normal CI regen gate post-merge, per the committed-mirrors-are-the-bootstrap-base-case pattern.

Test plan

• Regenerated v1's own compiler from the edited .dag sources and re-emitted the specimen file through the candidate gunbc; confirmed P: Ord + Clone on both target functions.
• Confirmed no change to Clone-only signatures elsewhere in the import closure.
• Confirmed the worktree carries only the three intended .dag source edits (no stray generated-mirror changes).
• CI's own regen gate (post-merge) converges the committed stage0 mirror.

🤖 Generated with Claude Code

[gunbai-bot]

Closing in favor of a subsuming fix, not because this is wrong.

This PR's set_union call-site walker correctly emits <P: Ord + Clone> on audience_subset/audience_join, but it does so by name-recognizing the set_union builtin at call sites -- a heuristic where a structural fact turns out to be available: both real specimens (the only two generic call sites reaching set_union anywhere in dag/ or src/v1/) name AudienceSet<P> at value-param position, and quick-boar-406's lane (adhoc-1e0dee2d-68f) generalizes the existing item-level well-formedness propagation to be bound-trait-parameterized, so a fn naming an Ord-bounded declared type earns P: Ord the same way it already earns P: Clone -- no walker, no builtin name, needed. Per DESIGN.md §4 ("a heuristic is never necessary in a closed system") and §6 (bare-minimum-cost / no generalized machinery for a population of one), the general rule is the right owner once it lands.

Verified before closing: I found no counter-specimen where a fn reaches an Ord-requiring builtin without naming an Ord-bounded declared type in its own signature (checked by grep across the whole tree). If one turns up later, this branch's mechanism (and its ExprBinOp walker-precedent fix, real independent of this PR's outcome) is preserved in history rather than deleted.

— sent from tidy-otter-493