refactor: move spark table parsing to directive scanner

Relocate spark table logic from the block processor to a dedicated
directive scanner. This allows spark tables to be treated as
`:md-table` directives and ensures they are processed in a second pass
after the content index is fully populated.

src/cli/completions/block-processor.ts

@@ -7,7 +7,6 @@
 
 import { posix } from "path";
 import { createHash } from "crypto";
-import Slugger from "github-slugger";
 import {
   parseStatYaml,
   parseStatCsv,
@@ -20,16 +19,14 @@ import {
   FENCED_BLOCK_RE,
   parseBlockAttrs,
   resolveBlockAs,
-  parseSparkTableCsv,
 } from "./block-scanner.js";
-import type { SparkTableCompletion, StatSheet } from "./types.js";
+import type { StatSheet } from "./types.js";
 
 // Re-export shared pieces for convenience
 export {
   FENCED_BLOCK_RE,
   parseBlockAttrs,
   resolveBlockAs,
-  parseSparkTableCsv,
   type BlockAttrs,
 } from "./block-scanner.js";
 
@@ -40,7 +37,6 @@ export {
 export interface ProcessedBlocks {
   stats: StatDef[];
   statTemplates: StatTemplate[];
-  sparkTables: SparkTableCompletion[];
   statSheets: StatSheet[];
 }
 
@@ -68,7 +64,9 @@ function contentHash(body: string): string {
  *
  * - Strips/replaces blocks based on `as`
  * - Injects `role=file` bodies into the content index
- * - Collects stat/template/spark-table blocks for completions
+ * - Collects stat/template blocks for completions
+ * - role=spark-table blocks are converted to :md-table directives
+ *   (spark table completions are collected later by the directive scanner)
  */
 export function processBlocks(
   content: string,
@@ -76,12 +74,10 @@ export function processBlocks(
   index: Record<string, string>,
 ): BlockResult {
   const fileDir = posix.dirname(fileRelativePath);
-  const slugger = new Slugger();
 
   const blocks: ProcessedBlocks = {
     stats: [],
     statTemplates: [],
-    sparkTables: [],
     statSheets: [],
   };
 
@@ -122,11 +118,6 @@ export function processBlocks(
         blocks.statTemplates.push(result.template);
       }
 
-      if (attrs.role === "spark-table") {
-        const parsed = parseSparkTableCsv(body, fileRelativePath, slugger);
-        if (parsed) blocks.sparkTables.push(parsed);
-      }
-
       if (attrs.role === "file") {
         const filename = attrs.id
           ? `${attrs.id}.${attrs.lang || "txt"}`