Add Markdown as an emit medium: row-driven serialize_markdown_source (medium rows)

[briansrls]

Markdown as an emit medium — the row-driven serialize_markdown_source

Adds Markdown as an emit target: the medium rows for headings, lists, GFM task-checkboxes, links, tables, code (inline + fenced), emphasis/strong, blockquotes — authored as a spelling-row table the fold looks up, never inline delimiter literals (the medium-as-Node guard, #5496).

Substrate decision (Path B, ruled by emission manager quick-seal-137)

A byte-faithful MarkdownDocument → markdown-SOURCE serializer on the v1 seed — the peer of extdeps/formats/yaml.dag serialize_yaml, and the byte-for-byte clone of the ci.yml pattern (serialize_yaml ∘ project_workflow_to_yaml). This is the near-term ROADMAP-inversion path that lands before self-host (plan invert-hand-maintained §7), and is SCAFFOLD-marked to dissolve into the eventual v2 06_translate Markdown TargetModel — the same spelling rows relocate, no rework.

Changes

• dsl/std/markdown.dag — extend the shared IR (single authority, §3): CheckboxState = Checked | Unchecked (a closed 2-sum, not Optional<Bool> — dodges the Value::Null literal-null straddle), MarkdownTaskItem, TaskListBlock, TableBlock, BlockquoteBlock. Adds the matching HTML-fragment arms + fail-closed XSS URL-scan arms so the existing IR→HTML path stays exhaustive and sound.
• dsl/extdeps/languages/markdown.dag — the MarkdownSpellings row table (CommonMark 0.31.2 + GFM 0.29 spellings, cited) + the row-driven serialize_markdown_source fold (and serialize_markdown_source_with(doc, spellings) for parameterization).
• dsl/test/claim/markdown_source_emit_test.dag — by-execution witness: exact bytes over a document exercising every construct family incl. nested task lists, plus a row-discipline tooth (perturbing the bullet row goes RED — verified false-on-perturb, true-on-revert).

Verification (by execution, v1 seed)

• markdown_source_emit_keystone_holds() → true
• existing markdown_markup_xss_keystone_holds() → true (HTML path unbroken)
• whole-tree gunbc compile --target rust: 408 files emitted, 0 diagnostics
• discriminating RED confirmed: a real bullet-row perturbation flips the witness to false

Consumer interface

serialize_markdown_source(doc: MarkdownDocument) -> String + std.markdown { CheckboxState, MarkdownTaskItem, TaskListBlock } — consumed by quiet-hawk-47's project_roadmap_to_markdown.

🤖 Generated with Claude Code

[gunbai-bot]

CI note for reviewers/operator — the failing per-head check is cross-repo, not this diff.

The modeling-coherence check fails inside the ctrl repo, not gunbc:
circular dependency: plans.coherence -> plans.reducible -> coherence.coherence_input -> coherence.coherence_query and module 'std.reducible' not found (imported by 'plans.reducible') at /home/briansrls/ctrl/plans/lib/reducible.dag.

This PR's diff is gunbc-only (three dsl/ markdown files) and touches zero reducible/coherence/plans modules — std.reducible exists fine in gunbc as dsl/std/reducible.dag. The breakage is a pre-existing ctrl/plans corpus issue gating all gunbc PR heads; it is not fixable from this PR (the ctrl repo isn't present in this session).

For the gunbc side: whole-tree gunbc compile --source-root dsl --source-root src/v2 --target rust → 408 files emitted, 0 diagnostics, and both witnesses are green by execution (markdown_source_emit_keystone_holds, existing markdown_markup_xss_keystone_holds). The one earlier dsl_compile_clean_gate failure was an intermediate auto-WIP snapshot (new MarkdownBlock variants committed before the matching HTML/XSS arms — a transient non-exhaustive match), resolved in the final commit.

— sent from deep-newt-165

[gunbai-bot]

Update — the gunbc ci failure was a real anchor-discipline defect in this diff (now fixed in e817836); correcting my earlier note.

When the gunbc CI floor finished it caught a genuine fail-closed I had missed pre-push: the new dsl/extdeps/languages/markdown.dag cited CommonMark/GFM only in comments and lacked the structured data extdeps_external_authority_anchor: ExternalAuthority that the external-authority lens (§3) requires of every extdeps/ module. Correctly fail-closed — a new extdeps module is neither machinery-exempt nor in the backfill_pending snapshot. My earlier comment under-ran the floor locally (witness + compile-clean only, not the live external-authority scan), so I missed it.

Fixed: added the anchor (spec.commonmark.org/0.31.2, below the imports, matching yaml.dag/elf). Re-verified locally: corpus_live_clean_tree_holds → true, markdown witness → true, whole-tree compile-clean 408 files / 0 diagnostics. New CI is running.

The separate modeling-coherence per-head check remains the cross-repo ctrl/plans breakage described above — unrelated to this diff.

— sent from deep-newt-165