docs(emission): regime-2 shared-fold design (collapse boutique *_emit serializers)

[gunbai-bot]

Summary

Design doc for the regime-2 emission generalization — collapsing the boutique pure-projection serializers (serialize_yaml, serialize_gitignore, serialize_runner_deploy) into one render(doc, protocol) fold over a shared std layout IR. This is the emit half that complements the gate unification (#5568, the commit/drift half).

Key framing the doc establishes:

• Two emission regimes: regime-1 (grammar-inverse, emission = ingestion⁻¹, round-trip-gated — languages; bright-stag's markdown lane) vs regime-2 (pure projection, emit-only, never parsed → no round-trip oracle, §5 honesty boundary).
• The boutique pain is entirely in regime-2: three hand-rolled folds re-inventing indent/line/separator layout over three different-or-absent IRs.
• The fix: a shared layout Doc IR + per-format Spelling protocol record + one render fold; expected_X = render(project_X_to_doc(authority), X_protocol). Discriminator that keeps it honest: zero format-specific layout branches in render.
• Faithfulness = byte-identical output (refactor, not behavior change); public entry-point signatures preserved so Unify the three per-artifact drift gates CiYamlGate RoadmapGate GitignoreGate into ONE universal generated-artifact policy - model the 3 questions generate-commit-drift with commit-policy DERIVED from the gitignore dag knowledge - registry-fold not opt-in - load-bearing gate infra, model-checkpoint- #5568's universal gate is unaffected.
• Scaffold-marked: dissolves into the v2 TargetModel grammar rows at self-host.

Docs-only; the worker authority for the dispatched build. std.markdown / roadmap_emit are explicitly out of scope (bright-stag's regime-1 lane).

Test plan

• Docs-only — no build/test impact.

🤖 Generated with Claude Code

[gunbai-bot]

CI failure at ba65dae is a runner-side sccache build flake, not a content issue: the job log shows Compilation failures 6 from sccache + the build-verification gate failing closed on an absent/stale gunbc/claim_executor artifact (the known DESIGN §5 fail-open guard catching cache corruption). This PR is docs-only — its code is byte-identical to main (1 file added, +109, 0 commits behind base 85a1e8b), so a compilation failure cannot originate here. Re-ran the failed job in place (same SHA, approval preserved). No fix commit is appropriate — there's nothing to fix. — sent from warm-lark-306

[briansrls]

CI failure at c336bdd is the sccache fleet-pressure build flake, not content: the job log shows Compilation failures 11, sccache: encountered fatal error / cargo/sccache EAGAIN under fleet pressure, and the build-verification gate failing closed on absent gunbc/claim_executor artifacts (the DESIGN §5 guard catching cache corruption). This PR is docs-only — its content vs main is a single design doc (+109, the head moved only because main was merged in to pick up #5574). A docs-only change cannot produce compilation failures. Re-ran the failed jobs in place (same SHA, approval preserved). No fix commit is appropriate. Note: the bot GitHub-API installation is also rate-limited (HTTP 403) right now — the fleet is under broad pressure (consistent with the host pin-skew + sccache flake I flagged to the operator). — sent from warm-lark-306