refactor: remove implicit content sniffing and table conversion

- Split loadCSV into parseCSVString (content) and loadCSVFromPath
  (path); drop isCSV/looksLikeCsv heuristics
- Delete coerceSparkTables and markedTable label-header magic;
  plain markdown tables now render as plain tables
- Add explicit markdown role=spark-table fence syntax that converts
  pipe tables to CSV at scan time, with dice-header validation
- Map ESM-only github-slugger and csv-parse browser build to CJS in
  jest config; add content-registry tests

src/cli/content-registry.ts

@@ -114,8 +114,8 @@ export interface DocScanResult {
 /**
  * Process a single markdown doc:
  *   - strips/replaces attributed fenced code blocks based on `as`
- *   - coerces spark-shaped markdown tables to `:md-table` directives
- *   - collects inline content (role=file, md-* bodies, spark tables, declare)
+ *   - converts `markdown`-lang `role=spark-table` bodies (pipe tables) to CSV
+ *   - collects inline content (role=file, md-* bodies, declare)
  *     into the doc's content store
  *
  * Does NOT touch the path index — the caller assembles the registry.
@@ -167,11 +167,15 @@ export function scanDoc(content: string, docPath: string): DocScanResult {
       }
 
       if (effectiveAs.startsWith("md-")) {
-        const id = deriveContentId("csv", body, attrs.id);
+        let blockBody = body;
+        if (attrs.role === "spark-table" && isMarkdownTableLang(attrs.lang)) {
+          blockBody = markdownTableBodyToCsv(body, docPath);
+        }
+        const id = deriveContentId("csv", blockBody, attrs.id);
         contentStore[id] = {
           id,
           kind: "csv",
-          body,
+          body: blockBody,
           role: attrs.role,
           as: effectiveAs,
         };
@@ -189,10 +193,7 @@ export function scanDoc(content: string, docPath: string): DocScanResult {
     },
   );
 
-  // ---- Pass 2: coerce spark-shaped markdown tables to :md-table ----
-  const rewritten = coerceSparkTables(stripped, contentStore);
-
-  return { stripped: rewritten, content: contentStore };
+  return { stripped, content: contentStore };
 }
 
 /**
@@ -241,8 +242,8 @@ export function buildRegistryFromIndex(
 
 /**
  * Whether a table header cell is a dice formula (a "spark table" first
- * column). Single source of truth shared by the CLI scanner and the frontend
- * `markedTable` renderer so both agree on what counts as a spark table.
+ * column). Used to validate `role=spark-table` blocks (CSV or markdown
+ * pipe-table bodies).
  */
 export function isSparkTableHeader(header: string): boolean {
   return /^\d*d\d+(?:[+-]\d+)?$/i.test(header.trim());
@@ -294,74 +295,43 @@ function markdownTableToCsv(
 }
 
 /**
- * Coerce spark-shaped markdown tables (first column header is a dice formula)
- * into `:md-table` directives, storing the CSV in the doc's content store and
- * injecting `data-spark` for the reveal feature.
+ * Languages whose fenced-block bodies are markdown pipe tables. Used with
+ * `role=spark-table` to convert the table to CSV at scan time — explicitly
+ * authorized by the role, never by content shape.
  */
-function coerceSparkTables(
-  content: string,
-  contentStore: Record<string, DocContent>,
-): string {
-  const mdTableRegex = /^(\|.+\|)\n(\|[-: |]+\|)\n((?:\|.+\|\n?)+)/gm;
+const MARKDOWN_TABLE_LANGS = new Set(["markdown", "md"]);
 
-  interface TableMatch {
-    fullMatch: string;
-    headerRow: string;
-    separatorRow: string;
-    bodyRowsText: string;
-    index: number;
-  }
-  const tableMatches: TableMatch[] = [];
+function isMarkdownTableLang(lang: string): boolean {
+  return MARKDOWN_TABLE_LANGS.has(lang.toLowerCase());
+}
 
-  let mdMatch: RegExpExecArray | null;
-  while ((mdMatch = mdTableRegex.exec(content)) !== null) {
-    const [, headerRow, separatorRow, bodyRowsText] = mdMatch;
-    const headers = splitTableRow(headerRow);
-    if (!isSparkTableHeader(headers[0])) continue;
-
-    const bodyRows = bodyRowsText
-      .trim()
-      .split(/\n/)
-      .filter((r) => r.trim().startsWith("|"));
-    const csv = markdownTableToCsv(headerRow, separatorRow, bodyRows);
-    if (!csv) continue;
-
-    tableMatches.push({
-      fullMatch: mdMatch[0],
-      headerRow,
-      separatorRow,
-      bodyRowsText,
-      index: mdMatch.index,
-    });
+/**
+ * Convert a fenced markdown pipe-table body to CSV for `role=spark-table`
+ * blocks. Validates the dice-formula first column (warning only — the role
+ * already declared intent) and falls back to storing the body as-is when it
+ * is not a recognizable pipe table.
+ */
+function markdownTableBodyToCsv(body: string, docPath: string): string {
+  const lines = body
+    .trim()
+    .split(/\r?\n/)
+    .filter((l) => l.trim().startsWith("|"));
+  if (lines.length < 2) {
+    console.warn(
+      `[content-registry] ${docPath}: role=spark-table markdown body is not a pipe table; storing as-is`,
+    );
+    return body;
   }
 
-  let rewritten = content;
-  for (let i = tableMatches.length - 1; i >= 0; i--) {
-    const m = tableMatches[i];
-    const bodyRows = m.bodyRowsText
-      .trim()
-      .split(/\n/)
-      .filter((r) => r.trim().startsWith("|"));
-    const csv = markdownTableToCsv(m.headerRow, m.separatorRow, bodyRows)!;
-
-    const id = deriveContentId("csv", csv);
-    contentStore[id] = {
-      id,
-      kind: "csv",
-      body: csv,
-      role: "spark-table",
-      as: "md-table",
-    };
-
-    const slug = sparkSlug(csv);
-    const directive = `:md-table[./${id}]{data-spark="${slug}"}`;
-    rewritten =
-      rewritten.slice(0, m.index) +
-      directive +
-      rewritten.slice(m.index + m.fullMatch.length);
+  const [headerRow, separatorRow, ...rows] = lines;
+  const headers = splitTableRow(headerRow);
+  if (!isSparkTableHeader(headers[0] || "")) {
+    console.warn(
+      `[content-registry] ${docPath}: spark table first column "${headers[0]}" is not a dice formula`,
+    );
   }
 
-  return rewritten;
+  return markdownTableToCsv(headerRow, separatorRow, rows) ?? body;
 }
 
 // ---------------------------------------------------------------------------
@@ -422,10 +392,12 @@ export function injectSparkDirectives(
 /**
  * Resolve a content reference from within a doc.
  *
- *   - `ref` is inline CSV → returned as-is.
  *   - `ref` is an absolute path → path index.
  *   - `ref` is a relative path → resolved against the doc directory; checks
- *     the path index first, then the doc's inline content store.
+ *     the doc's inline content store first, then the path index.
+ *
+ * Refs are always ids or paths — inline content must be defined in a fenced
+ * block and referenced by its `./{id}`.
  *
  * Returns `null` when nothing matches.
  */
@@ -437,9 +409,6 @@ export function resolveContent(
   const trimmed = ref.trim();
   if (!trimmed) return null;
 
-  // Inline CSV body.
-  if (looksLikeCsv(trimmed)) return trimmed;
-
   if (trimmed.startsWith("/")) {
     return registry.pathIndex[trimmed] ?? null;
   }
@@ -477,25 +446,6 @@ export function resolveInlineByPath(
   return null;
 }
 
-/** Naive CSV sniff — matches the frontend `isCSV` heuristic. */
-function looksLikeCsv(str: string): boolean {
-  const trimmed = str.trim();
-  if (trimmed.startsWith("---\n") || trimmed.startsWith("---\r\n")) return true;
-
-  const lines = trimmed.split(/\r?\n/).filter((line) => line.trim() !== "");
-  if (lines.length < 2) return false;
-
-  const separators = [",", "\t", ";", "|"];
-  const firstLine = lines[0];
-  for (const sep of separators) {
-    if (firstLine.includes(sep)) {
-      const hasInOthers = lines.slice(1).some((line) => line.includes(sep));
-      if (hasInOthers) return true;
-    }
-  }
-  return false;
-}
-
 // ---------------------------------------------------------------------------
 // Derived completions
 // ---------------------------------------------------------------------------