Potential memory leak in docling pdf conversion

### Bug
I've found memory use gradually growing with multiple successive conversions of simple PDF documents using the standard PDF conversion pipeline. I have observed this behaviour with/without OCR enabled and when switching to PyPdfium backend also.

I have also not found any way to free this memory, deleting and garbage collecting all references do not seem to clear it. Instantiating a new DocumentConverter also just increases the base memory load (I believe it must reload a lot of the heavy ML stuff in the background)

### Steps to reproduce
I can reproduce the issue with this simple script:

```python
from docling.document_converter import DocumentConverter
from pathlib import Path
import psutil
import os

def get_current_process_memory() -> float:
"""Prints current process memory usage."""
process = psutil.Process(os.getpid())
mem_mb = process.memory_info().rss / 1024 / 1024
print(f"Total Memory: {mem_mb:.2f} MB")
return mem_mb

def plot_memory_usage() -> None:
import pandas as pd
import matplotlib.pyplot as plt

df = pd.DataFrame(memory_usage, columns=["memory_usage"])
df.plot(kind="line")
# save plot to png
plt.savefig("memory_usage.png")

if __name__ == "__main__":
converter = DocumentConverter()
pdf = Path("lorem_ipsum.pdf")
memory_usage = []

for i in range(100):
doc = converter.convert(pdf)
memory_usage.append(get_current_process_memory())

plot_memory_usage()
```

This is using this simple 1 page sample PDF

[lorem_ipsum.pdf](https://github.com/user-attachments/files/24164895/lorem_ipsum.pdf)

And produces the following output graph of memory use (X axis: iterations, Y axis: memory (MB)):

### Docling version
Docling version: 2.64.0
Docling Core version: 2.54.0
Docling IBM Models version: 3.10.3
Docling Parse version: 4.7.2
Python: cpython-312 (3.12.12)
Platform: Windows-11-10.0.26100-SP0

### Python version
3.12.12