Short isolated paragraphs dropped from extraction — layout model fails to detect small text blocks

### Bug
Docling silently drops short, isolated single-line paragraphs from PDF text extraction. The text is visually present in the PDF and correctly extractable by `pymupdf`'s `get_text()`, but does not appear in Docling's output.

The specific line dropped: `UPDATE: Nedim testing the sync!!!` - a single short paragraph between a multi-line intro paragraph and a section heading.

This was tested with three different PDF variants:
- **Original PDF** (Type3 fonts from an HTML-to-PDF converter): line missing
- **Rebuilt PDF with standard Helvetica fonts** (via pymupdf): line missing
- **Image-only PDF** (rendered at 200 DPI, no text layer, forcing OCR): line missing

All three variants produce the same visual page, suggesting the issue is in the **layout detection model** (object detection on rendered images), not in text-layer parsing or font handling.

Pipeline options used:
```python
pipeline_options = PdfPipelineOptions(
do_ocr=True,
ocr_options=TesseractCliOcrOptions(lang=["eng"]),
do_table_structure=True,
)
```

Backend: PyPdfiumDocumentBackend

### Steps to reproduce
1. Use the attached `file_with_dropped_line.pdf`
2. Extract with Docling using default PDF pipeline with do_ocr=True
3. Check the output - the line "UPDATE: Nedim testing the sync!!!" on page 1 (between the intro paragraph and "Section 1: Access Control Requirements") is missing from all output
4. Verify with pymupdf: fitz.open("file.pdf")[0].get_text() - the line IS present

### File

[file_with_dropped_line.pdf](https://github.com/user-attachments/files/26624169/file_with_dropped_line.pdf)

### Docling version
2.64.0

### Python version
3.13