[Bee] Excel backend duplicates cells of a sparse table as extra tables

Bug

When a worksheet has a single table with empty cells, non-empty cells that are not directly adjacent to the rest of the table (e.g. a column whose first value has an empty cell above it) are exported twice: once inside the main table, and again as a separate small table.

Minimal sheet (sparse_sheet.xlsx): sparse_sheet.xlsx

Expected: 1 table of shape (4, 3). Actual: 2 tables, shapes [(4, 3), (2, 1)]. Markdown export:

|   ID | Name   | Note   |
|------|--------|--------|
|    1 | alpha  |        |
|    2 |        | foo    |
|    3 |        | bar    |

| foo   |
|-------|
| bar   |

foo/bar already appear in the first table and are repeated as a standalone table. On real sparse sheets (e.g. ~100-column exports) this produces many such duplicate fragment tables.

Steps to reproduce

Attached: sparse_sheet.xlsx and the code:

from pathlib import Path
from docling.document_converter import DocumentConverter

xlsx = Path("sparse_sheet.xlsx")
doc = DocumentConverter().convert(xlsx).document
xlsx.with_suffix(".md").write_text(doc.export_to_markdown())

shapes = [(t.data.num_rows, t.data.num_cols) for t in doc.tables]
print(f"actual: {len(doc.tables)} tables of shapes {shapes}")
assert shapes == [(4, 3)]

Output:

actual: 2 tables of shapes [(4, 3), (2, 1)]
AssertionError

Docling version

Docling version: 2.110.0
Docling Core version: 2.91.0
Docling IBM Models version: 3.14.0

Python version

Python 3.12.12