R3 gate #47: auto loop parallelism unproven falls back sequential

[briansrls]

Auto-opened by session-dashboard for session sharp-wolf-424.

Summary

Implements R3 gate #47 (auto_loop_parallelism_unproven_falls_back_sequential). The TestClaim uses LensOutputEquals with r3_auto_loop_parallelism_sequential_emit_witness, which checks emit_rust output does not take the parallel batch path (std::thread::scope), consistent with the gate #43 structural proxy for bind parallelism.

INVARIANTS P5 — single checkable receipt

Per INVARIANTS.md § P5 / Dispatch-Discipline (b) Per-PR gate, this PR states exactly one checkable receipt: an explicit deferral that names a concrete ROADMAP.md row — DB-20 under § Active deferrals (ROADMAP.md on main, bullet ~L318). Dissolution trigger for the hand-Rust substring proxy: when the DB-20 / ordinary-lens workflow parallelism surface replaces whole-program thread::scope greps — tracked in-tree by TODO(DB-20) next to the gate #47 LensOutputEquals arm in src/v3/compiler/src/test_runner.rs. No SG-0 census net shrink in this PR (extends existing witness dispatch only).

Supporting links: docs/db-history/db-20.md, docs/design-db20-lane2-stage2e-parallelism-lens.md.

Test plan

• CTRL_BUILD_BYPASS_SHIMS=1 cargo test -p v3-compiler --test integration r3_free_consequences_second_batch_reaches_expected_consumer_shapes
• CTRL_BUILD_BYPASS_SHIMS=1 cargo test -p v3-compiler --test integration r3_free_consequences_first_batch_reaches_unified_predicate_shape
• cargo fmt --all --check
• CTRL_BUILD_BYPASS_SHIMS=1 cargo clippy -p v3-compiler --all-targets -- -D warnings

Worker attestation

• Title describes the change (not the session id or branch).
• PR body summarises what and why (above).
• Tests run: commands listed; matching results on push branch.
• No Closes #N — dashboard work item closes on merge via session linkage.
• No commits on this branch are surprises.
• No secrets / credentials / large binaries staged.

[briansrls]

Review follow-up (claude opus APPROVE_WITH_COMMENTS): Landed in 2778c9a48 — rustdoc + inline comments now cite the named deferral receipt explicitly: ROADMAP.md § Active deferrals → DB-20, docs/db-history/db-20.md, and docs/design-db20-lane2-stage2e-parallelism-lens.md. PR description updated with the same. Added TODO(DB-20) next to the gate #47 arm for replacing whole-program thread::scope substring checks with a scheduling-local marker when the producer lands. Leaving the parallel/sequential dispatch arms unmerged until a third witness appears (per exploratory note).

[briansrls]

Dashboard routing note: The quoted thread is the session’s own prior follow-up (2778c9a48 recap), not a new reviewer finding — nothing further to implement.

Code verification: On main’s branch tip, test_runner.rs still carries the DB-20 deferral citations and TODO(DB-20) adjacent to the gate #47 LensOutputEquals arm (same intent as that comment).

Merge readiness (this environment): fmt + ci are green; v3 is still pending on the latest run, so CI is not fully green yet from here. I also cannot enumerate Verdict: APPROVE lines from api-review via gh in this sandbox — please treat dashboard-ops reviews gunb-ai/gunbc#2532 as authoritative for the ≥2-approval rule before squash-merge.

— sent from sharp-wolf-424

[briansrls]

composer-2 (8755): Verdict APPROVE acknowledged. No blocking findings; exploratory substring-proxy limitation already covered by TODO(DB-20) and matches gate #43 witness style in the same module — no further change.

— sent from sharp-wolf-424