XLSX embedded images are missing from Document.assets

## Summary

`anydoc.to_document(..., "xlsx")` extracts worksheet cell content but omits embedded worksheet images from `Document.assets`.

This differs from #63: that issue reports assets being present for DOCX but absent from rendered Markdown. In this case, the XLSX asset is already missing from the shared `Document` model, so consumers cannot preserve the image bytes or create their own Markdown reference.

## Environment

- `firecrawl-anydoc`: 0.2.3
- Python: 3.11.15
- OS: Windows

## Minimal reproduction

```python
import tempfile
from pathlib import Path

import anydoc
from openpyxl import Workbook
from openpyxl.drawing.image import Image

PNG = bytes.fromhex(
"89504e470d0a1a0a0000000d49484452000000010000000108060000001f15c489"
"0000000d49444154789c6360f8cfc000000301010018dd8db10000000049454e44"
"ae426082"
)

with tempfile.TemporaryDirectory() as directory:
root = Path(directory)
image_path = root / "evidence.png"
workbook_path = root / "with-image.xlsx"
image_path.write_bytes(PNG)

workbook = Workbook()
worksheet = workbook.active
worksheet["A1"] = "visual evidence"
worksheet.add_image(Image(str(image_path)), "B2")
workbook.save(workbook_path)
workbook.close()

payload = workbook_path.read_bytes()
document = anydoc.to_document(payload, "xlsx")

print("format:", anydoc.format_from_path(workbook_path))
print("assets:", len(document.assets))
print("blocks:", len(document.blocks))
print(anydoc.to_markdown_bytes(payload, "xlsx"))
```

## Actual result

```text
format: xlsx
assets: 0
blocks: 1
| |
| --- |
| visual evidence |
```

The cell content is extracted, but the embedded PNG is absent from `document.assets`.

## Expected result

`document.assets` should contain the embedded worksheet image, including its bytes and media type, consistently with the documented shared document model contract for embedded images.

Positional image rendering in Markdown can remain tracked separately in #63; exposing the image in `Document.assets` is sufficient for consumers to persist it themselves.

Start by reproducing the XLSX case through anydoc.to_document with the embedded PNG shown in the issue, then trace the XLSX conversion path that populates Document.assets. Done means the extracted image appears in document.assets with its original bytes and media type, while the existing worksheet cell content remains available.

Written by the indexing model from the issue text.