fix(ocr-pipeline): stop noise filter from stripping parenthesized words

_is_noise_tail_token() treated words with unbalanced parentheses like
"selbst)" or "(wir" as OCR noise because the parenthesis counted as
"internal noise". Now strips leading/trailing parentheses before the
noise check, so legitimate words in example sentences like
"We baked ... (wir ... selbst)" are preserved.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

klausur-service/backend/cv_vocab_pipeline.py

@@ -3289,19 +3289,22 @@ def _is_noise_tail_token(token: str) -> bool:
     t_check = stripped_punct if stripped_punct else t
 
     # Check for legitimate punctuation patterns vs. real noise.
-    # Legitimate: "(auf)", "under-", "e.g.", "(on)"
+    # Legitimate: "(auf)", "under-", "e.g.", "(on)", "selbst)", "(wir"
     # Noise: "Es)", "3d", "B|"
-    # Strategy: strip balanced parens & trailing hyphens, THEN check residual.
+    # Strategy: strip parentheses & trailing hyphens, THEN check residual.
     t_inner = t_check
     # Remove balanced parentheses wrapping the token: "(auf)" → "auf"
     if t_inner.startswith('(') and t_inner.endswith(')'):
         t_inner = t_inner[1:-1]
+    # Remove unbalanced parentheses at start/end (common in example sentences):
+    # "(wir" → "wir", "selbst)" → "selbst", "(selbst))" → "selbst"
+    t_inner = t_inner.lstrip('(').rstrip(')')
     # Remove trailing hyphen (word continuation): "under-" → "under"
     if t_inner.endswith('-'):
         t_inner = t_inner[:-1]
     # Now check: does the inner form still have non-alpha noise?
     inner_alpha = ''.join(_RE_ALPHA.findall(t_inner))
-    has_internal_noise = len(t_inner) > len(inner_alpha)
+    has_internal_noise = len(t_inner) > len(inner_alpha) if t_inner else False
 
     # Long alpha words (4+ chars) without internal noise are likely real
     if len(cleaned) >= 4 and not has_internal_noise: