Add merged-word splitting to OCR spell review

OCR often merges adjacent words when spacing is tight, e.g.
"atmyschool" → "at my school", "goodidea" → "good idea".

New _try_split_merged_word() uses dynamic programming to find the
shortest sequence of dictionary words covering the token. Integrated
as step 5 in _spell_fix_token() after general spell correction.

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

klausur-service/backend/cv_review.py

@@ -720,6 +720,58 @@ def _spell_dict_knows(word: str) -> bool:
     return bool(_en_spell.known([w])) or bool(_de_spell.known([w]))
 
 
+def _try_split_merged_word(token: str) -> Optional[str]:
+    """Try to split a merged word like 'atmyschool' into 'at my school'.
+
+    Uses dynamic programming to find the shortest sequence of dictionary
+    words that covers the entire token.  Only returns a result when the
+    split produces at least 2 words and ALL parts are known dictionary words.
+
+    Preserves original capitalisation by mapping back to the input string.
+    """
+    if not _SPELL_AVAILABLE or len(token) < 5:
+        return None
+
+    lower = token.lower()
+    n = len(lower)
+
+    # dp[i] = shortest list of word lengths that covers lower[:i], or None
+    dp: list = [None] * (n + 1)
+    dp[0] = []
+
+    for i in range(1, n + 1):
+        # Try all possible last-word lengths (2..min(i, 20))
+        # Allow single-char words only for 'a' and 'I'
+        min_len = 1
+        for j in range(max(0, i - 20), i):
+            if dp[j] is None:
+                continue
+            word_len = i - j
+            candidate = lower[j:i]
+            if word_len == 1 and candidate not in ('a', 'i'):
+                continue
+            if word_len < 2 and candidate not in ('a', 'i'):
+                continue
+            if _spell_dict_knows(candidate):
+                new_split = dp[j] + [word_len]
+                # Prefer fewer words (shorter split)
+                if dp[i] is None or len(new_split) < len(dp[i]):
+                    dp[i] = new_split
+
+    if dp[n] is None or len(dp[n]) < 2:
+        return None
+
+    # Reconstruct with original casing
+    result = []
+    pos = 0
+    for wlen in dp[n]:
+        result.append(token[pos:pos + wlen])
+        pos += wlen
+
+    logger.debug("Split merged word: %r → %r", token, " ".join(result))
+    return " ".join(result)
+
+
 def _spell_fix_token(token: str, field: str = "") -> Optional[str]:
     """Return corrected form of token, or None if no fix needed/possible.
 
@@ -777,6 +829,14 @@ def _spell_fix_token(token: str, field: str = "") -> Optional[str]:
                     correction = correction[0].upper() + correction[1:]
                 if _spell_dict_knows(correction):
                     return correction
+
+    # 5. Merged-word split: OCR often merges adjacent words when spacing
+    #    is too tight, e.g. "atmyschool" → "at my school"
+    if len(token) >= 5 and token.isalpha():
+        split = _try_split_merged_word(token)
+        if split:
+            return split
+
     return None