fix(vocabService): skip multi-word learning items in SRS item processing

backend/services/vocabService.js

@@ -2672,6 +2672,11 @@ export default class VocabService {
     let maxNum = list.reduce((m, ex) => Math.max(m, Number(ex.exerciseNumber) || 0), 0);
     const augmentedRows = rows.map((r) => ({ ...r, allReferences }));
     for (const row of augmentedRows) {
+      // Skip multi-word learning items (they are sentences, not single lexemes)
+      const learningText = String(row.learning || '').trim();
+      if (!learningText || learningText.split(/\s+/).length > 1) {
+        continue;
+      }
       if (this._lexemePairCoveredByMultipleChoice(list, row.learning, row.reference)) {
         continue;
       }