Documents output filled with GLYPH word

### Bug
Docling sometimes produces unreadable output after converting document. For example
(force_full_page_ocr set to False)
```
## PRZEDMIAR ROBÓT

!GLYPH<15>GLYPH<20>GLYPH<22>'()GLYPH<23>GLYPH<20>GLYPH<18>*GLYPH<20>GLYPH<4>GLYPH<16>GLYPH<17>+,-GLYPH<4>GLYPH<21>.GLYPH<13>GLYPH<4>GLYPH<8>GLYPH<22>/,GLYPH<15>01.GLYPH<17>GLYPH<4>"GLYPH<19>GLYPH<17>GLYPH<21>0)GLYPH<23>GLYPH<20>GLYPH<4>#GLYPH<20>2,GLYPH<21>)13
```
(force_full_page_ocr set to True)
```
- " wykonanie tynkéw, tynkéw uzupetniajacych, posadzek, oktadzin, malowanie,
- "montaz nowej stolarki okiennej i drzwiowej zwiazanej z rozbudowa,
- "uzupeinienia remontowe na elewacjach z malowaniem.

| GLYPH<31>/GLYPH<13>
| GLYPH<1>GLYPH<17>5'GLYPH<18>*1GLYPH<4> GLYPH<23>GLYPH<17>GLYPH<22>5-GLYPH<17>GLYPH<16>'GLYPH<22>GLYPH<17>GLYPH<28> GLYPH<21>1
GLYPH<20>5GLYPH<21>GLYPH<20>
```

Related issue: https://github.com/docling-project/docling/issues/185
`OcrOptions.force_full_page_ocr` helps a little, but not completely. With this option it breaks when it tries to parse table

[Example, not confidential document](https://github.com/user-attachments/files/20672460/przedmiary.pdf)
with force_full_page_ocr=True it breaks after 3rd page (inclusive)
...

### Steps to reproduce
```
pipeline_options = PdfPipelineOptions()
accelerator_options = AcceleratorOptions(num_threads=8, device=AcceleratorDevice.CPU)
pipeline_options.accelerator_options = accelerator_options

pipeline_options.do_ocr = True
pipeline_options.do_table_structure = True
pipeline_options.table_structure_options.do_cell_matching = True
pipeline_options.allow_external_plugins = True
pipeline_options.generate_page_images = True
pipeline_options.generate_picture_images = True
pipeline_options.do_picture_description = False
settings.debug.profile_pipeline_timings = True
ocr_options = TesseractCliOcrOptions(force_full_page_ocr=True)
pipeline_options.ocr_options = ocr_options

formatter_option = PdfFormatOption(
pipeline_options=pipeline_options,
)

converter = DocumentConverter(
format_options={
InputFormat.PDF: formatter_option,
InputFormat.IMAGE: formatter_option # Docling uses same pipeline for PDF and Images, so it's okay
}
)

doc = converter.convert("./data/examples/przedmiary.pdf").document
print(doc.export_to_markdown())
```
This code uses TesseractOCR, but the bug occurs even if I deliver complete, proper OCR result from external source, so it's not OCR bug

### Docling version
```
Docling version: 2.33.0
Docling Core version: 2.31.1
Docling IBM Models version: 3.4.3
Docling Parse version: 4.0.1
Python: cpython-312 (3.12.9)
Platform: Linux-6.13.10-200.fc41.x86_64-x86_64-with-glibc2.41
```

### Python version
`Python 3.12.9`

Other issues say that changing PDF backend might help, but on the same time they claim this change breaks other files