Weird ocr markown text inlining

So we're trying to inline ocr result inside markdown export.

```python
from docling.datamodel.accelerator_options import AcceleratorOptions, AcceleratorDevice
from docling.datamodel.base_models import InputFormat, DocumentStream
from docling.datamodel.pipeline_options import PdfPipelineOptions
from docling.document_converter import DocumentConverter, PdfFormatOption, PowerpointFormatOption, HTMLFormatOption, CsvFormatOption, ExcelFormatOption, MarkdownFormatOption, WordFormatOption
from pathlib import Path
from typing import Union
from docling_core.types.doc.base import ImageRefMode
from docling_core.types.doc import TextItem
class DoclingWorker:
_instance = None

def __new__(cls, *args, **kwargs):
if cls._instance is None:
cls._instance = super(DoclingWorker, cls).__new__(cls)
return cls._instance

def __init__(self, pipeline_options: PdfPipelineOptions = None):
if hasattr(self, "_initialized") and self._initialized:
return

if pipeline_options is None:
accelerator_options = AcceleratorOptions(
num_threads=1, device=AcceleratorDevice.CUDA, cuda_use_flash_attention2=True
)

pipeline_options = PdfPipelineOptions()
pipeline_options.do_ocr = False
pipeline_options.do_table_structure = True
pipeline_options.accelerator_options = accelerator_options

self.format_options = {
InputFormat.PDF: PdfFormatOption(pipeline_options=pipeline_options, generate_picture_images=True),
InputFormat.DOCX: WordFormatOption(pipeline_options=pipeline_options),
InputFormat.PPTX: PowerpointFormatOption(pipeline_options=pipeline_options),
InputFormat.HTML: HTMLFormatOption(pipeline_options=pipeline_options),
# InputFormat.ASCIIDOC: AsciiDocFormatOption(pipeline_options=pipeline_options),
InputFormat.MD: MarkdownFormatOption(pipeline_options=pipeline_options),
InputFormat.CSV: CsvFormatOption(pipeline_options=pipeline_options),
InputFormat.XLSX: ExcelFormatOption(pipeline_options=pipeline_options)
}
self.converter = DocumentConverter(
format_options=self.format_options
)
self._initialized = True

def convert_to_markdown(self, document:Union[Path, str, DocumentStream]) -> str:
result = self.converter.convert(document)

return result.document.export_to_markdown(image_mode=ImageRefMode.INLINE)

@property
def supported_formats(self):
return [key.value for key in self.format_options.keys()]

if __name__ == "__main__":
from docling.datamodel.pipeline_options import EasyOcrOptions
accelerator_options = AcceleratorOptions(
num_threads=1, device=AcceleratorDevice.CUDA, cuda_use_flash_attention2=True
)

pipeline_options = PdfPipelineOptions()
pipeline_options.do_ocr = True
pipeline_options.do_table_structure = True
pipeline_options.accelerator_options = accelerator_options

pipeline_options.force_backend_text = True

pipeline_options.ocr_options = EasyOcrOptions(
lang=['fr', 'en'],
bitmap_area_threshold=0.0,
confidence_threshold=0.0
)

pipeline_options.do_picture_description = False

dw = DoclingWorker(pipeline_options=pipeline_options)

files = ["./pdftest.pdf", "Coucou text du haut.pdf"]
for file in files:
print(f"--- Converting {file} ---")
md = dw.convert_to_markdown(file)
print(md)
print("--- End of Conversion ---\n\n")

```

So i have a try with two PDFS :

[Coucou text du haut.pdf](https://github.com/user-attachments/files/22490677/Coucou.text.du.haut.pdf)
[pdftest.pdf](https://github.com/user-attachments/files/22490676/pdftest.pdf)

pdftest looks like this :

But returns :

```text
Hey this is upper text

Hey this is bottom text

## THIS I5 AN IMAGE
```

Please note the image is placed at bottom

------------------------------

And with second pdf it returns :
```text
Coucou text du haut

Coucou texte du bas
```

When I apply the PR : https://github.com/docling-project/docling-core/pull/390

Second PDF turns into :

```text
Coucou text du haut

AEXTINCTION NON CONTROLÉE DE LA LUMIERE
RISQUES
DE CHUTES
UTILISATION D'APPAREIL DE VISION NOCTURNE
FORTEMENT CONSEILLÉ

Coucou texte du bas
```

Which is exactly (or kind of) what we expect.

1. OCR markdown export doesn't work on all pdfs/images
2. export isn't correctly placed (unless my fix is applied, but only applies on non accepted images)