fix(ocr): apply CJK filter to string output and sync with local improvements

ocr_module.py

@@ -55,7 +55,7 @@ def clean_latex_output(text: str) -> str:
 def extract_latex_from_pix2text(out) -> str:
     """Safely extract LaTeX text from pix2text output regardless of return type."""
     if isinstance(out, str):
-        return out
+        return clean_latex_output(out)
     elif isinstance(out, list):
         parts = []
         for item in out: