Add SmartSpellChecker + refactor vocab-worksheet page.tsx

SmartSpellChecker (klausur-service):
- Language-aware OCR post-correction without LLMs
- Dual-dictionary heuristic for EN/DE language detection
- Context-based a/I disambiguation via bigram lookup
- Multi-digit substitution (sch00l→school)
- Cross-language guard (don't false-correct DE words in EN column)
- Umlaut correction (Schuler→Schüler, uber→über)
- Integrated into spell_review_entries_sync() pipeline
- 31 tests, 9ms/100 corrections

Vocab-worksheet refactoring (studio-v2):
- Split 2337-line page.tsx into 14 files
- Custom hook useVocabWorksheet.ts (all state + logic)
- 9 components in components/ directory
- types.ts, constants.ts for shared definitions

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

klausur-service/backend/cv_review.py

@@ -881,10 +881,25 @@ def spell_review_entries_sync(entries: List[Dict]) -> Dict:
     """Rule-based OCR correction: spell-checker + structural heuristics.
 
     Deterministic — never translates, never touches IPA, never hallucinates.
+    Uses SmartSpellChecker for language-aware corrections with context-based
+    disambiguation (a/I), multi-digit substitution, and cross-language guard.
     """
     t0 = time.time()
     changes: List[Dict] = []
     all_corrected: List[Dict] = []
+
+    # Use SmartSpellChecker if available, fall back to legacy _spell_fix_field
+    _smart = None
+    try:
+        from smart_spell import SmartSpellChecker
+        _smart = SmartSpellChecker()
+        logger.debug("spell_review: using SmartSpellChecker")
+    except Exception:
+        logger.debug("spell_review: SmartSpellChecker not available, using legacy")
+
+    # Map field names → language codes for SmartSpellChecker
+    _LANG_MAP = {"english": "en", "german": "de", "example": "auto"}
+
     for i, entry in enumerate(entries):
         e = dict(entry)
         # Page-ref normalization (always, regardless of review status)
@@ -907,9 +922,18 @@ def spell_review_entries_sync(entries: List[Dict]) -> Dict:
             old_val = (e.get(field_name) or "").strip()
             if not old_val:
                 continue
-            # example field is mixed-language — try German first (for umlauts)
-            lang = "german" if field_name in ("german", "example") else "english"
-            new_val, was_changed = _spell_fix_field(old_val, field=lang)
+
+            if _smart:
+                # SmartSpellChecker path — language-aware, context-based
+                lang_code = _LANG_MAP.get(field_name, "en")
+                result = _smart.correct_text(old_val, lang=lang_code)
+                new_val = result.corrected
+                was_changed = result.changed
+            else:
+                # Legacy path
+                lang = "german" if field_name in ("german", "example") else "english"
+                new_val, was_changed = _spell_fix_field(old_val, field=lang)
+
             if was_changed and new_val != old_val:
                 changes.append({
                     "row_index": e.get("row_index", i),
@@ -921,12 +945,13 @@ def spell_review_entries_sync(entries: List[Dict]) -> Dict:
                 e["llm_corrected"] = True
         all_corrected.append(e)
     duration_ms = int((time.time() - t0) * 1000)
+    model_name = "smart-spell-checker" if _smart else "spell-checker"
     return {
         "entries_original": entries,
         "entries_corrected": all_corrected,
         "changes": changes,
         "skipped_count": 0,
-        "model_used": "spell-checker",
+        "model_used": model_name,
         "duration_ms": duration_ms,
     }