fix(training): only count reviewed blocks as checked text for recognition

Previously all MANUAL blocks counted as eligible training data, even ones
where text was filled in by guided OCR but never explicitly reviewed. This
caused segmentation and recognition counts to always match.

Now only reviewed=true blocks qualify for recognition training, so the
counts properly reflect: segments = all drawn annotation boxes,
checked text = only boxes where the user has verified the transcription.

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

backend/src/main/java/org/raddatz/familienarchiv/repository/TranscriptionBlockRepository.java

@@ -22,7 +22,7 @@ public interface TranscriptionBlockRepository extends JpaRepository<Transcriptio
             SELECT b FROM TranscriptionBlock b
             JOIN DocumentAnnotation a ON a.id = b.annotationId
             JOIN Document d ON d.id = b.documentId
-            WHERE (b.source = 'MANUAL' OR (b.source = 'OCR' AND b.reviewed = true))
+            WHERE b.reviewed = true
             AND 'KURRENT_RECOGNITION' MEMBER OF d.trainingLabels
             """)
     List<TranscriptionBlock> findEligibleKurrentBlocks();