Scanned PDF text turns into meaningless letters

### Bug

When using docling to parse scanned PDFs, it was found that some paragraphs turned into meaningless letters

### Steps to reproduce

The document parsing engine I set up in OpenWebUI is Docling. When interacting with the large language model, I uploaded a scanned PDF, and after parsing, I found that some paragraphs contained meaningless strings of letters like 'n sns s s a n'.
The original text of my file is as follows：

The result after docling parsing is as follows.

### Version
docling​ 2.63.0
Python​ 3.12

Reproduce the issue with the reported scanned PDF using Docling 2.63.0 and Python 3.12, then compare the parsed paragraphs with the source text shown in the attachments. Trace the document-parsing path that produces strings such as 'n sns s s a n'; done means the affected paragraphs retain meaningful text.

Written by the indexing model from the issue text.