[Bee] Hybrid PDF: top line words from embedded image are missing when the image is very close to the programmatic words

### Bug
When processing the hybrid pdf, docling will do cell merging between words from ocr and words from pdf parser. When the embedded image are very close to the programmatic words, the first line will got lost in the md or plain text output.

For example:

actually, the embedded image is not overlapped with programmatic words, but the first line is missing.
[hybrid_paragraph.pdf](https://github.com/user-attachments/files/26499730/hybrid_paragraph.pdf)

### Steps to reproduce
use default pdf pipeline settings with do_ocr enabled
...

### Docling version
2.73.1
...

### Python version
Python 3.11.10