fix: sva now extracts and reads properly

scripts/seed-sva.ts

@@ -93,24 +93,6 @@ async function insertBatched(
   for (let i = 0; i < docs.length; i++) {
     const sanitized = sanitizeDoc(docs[i]) as Record<string, unknown>;
 
-    // Work around BSON driver bug with large string arrays:
-    // Store paragraphs as a single JSON-encoded string instead of a native
-    // array. The data layer (lib/sva.ts) already handles stringified JSON
-    // via normalizeParagraphs().
-    if (Array.isArray(sanitized.paragraphs)) {
-      // Stringify, then sanitize the resulting string to ensure clean UTF-8
-      let jsonStr = JSON.stringify(sanitized.paragraphs);
-      // NFC normalize the whole JSON string
-      jsonStr = jsonStr.normalize("NFC");
-      // Remove any stray control chars or surrogates
-      jsonStr = jsonStr.replace(/[\x00-\x08\x0b\x0c\x0e-\x1f]/g, "");
-      jsonStr = jsonStr.replace(/[\uD800-\uDFFF]/g, "");
-      jsonStr = jsonStr.replace(/[\u200B-\u200F\u2028-\u202F\uFEFF]/g, "");
-      // Round-trip through Buffer to guarantee valid UTF-8 byte sequence
-      jsonStr = Buffer.from(jsonStr, "utf-8").toString("utf-8");
-      sanitized.paragraphs = jsonStr;
-    }
-
     try {
       await col.insertOne(sanitized);
     } catch (err) {