bug/XLSX Table string representation loses row boundaries after 0.16.0 causing flattened output from partition_xlsx

Describe the bug

After upgrading unstructured from 0.16.0 to 0.16.1 and later, partition_xlsx returns a Table element whose string representation no longer preserves stable row boundaries.

In 0.16.0, str(Table) allowed deterministic reconstruction of header and row groupings by splitting on double newlines.

Starting in 0.16.1, the same XLSX file produces a flattened string where rows are no longer clearly separated, which breaks header to value reconstruction logic that previously worked reliably.

This change appears related to the whitespace or HTML minification updates introduced around 0.16.1.

To Reproduce

Install two versions separately and run the same code against the same XLSX file.

from unstructured.partition.xlsx import partition_xlsx
from unstructured.documents.elements import Table
import io

def inspect_xlsx(path):
    with open(path, "rb") as f:
        file_content = io.BytesIO(f.read())

    elements = partition_xlsx(file=file_content)

    print("Element count:", len(elements))
    for i, el in enumerate(elements):
        print("Type:", type(el))
        if isinstance(el, Table):
            print("---- STRING OUTPUT START ----")
            print(str(el))
            print("---- STRING OUTPUT END ----")

inspect_xlsx("sample.xlsx")

Observed difference:

0.16.0
Row groupings are preserved in a way that allows:

rows = str(table).split("\n\n")

to produce logical rows.

0.16.1 and later
str(table) appears flattened, producing a single continuous block such as:

1 1 0 www.example.com 2 2 0 www.example2.com ...

Double newline row separation is no longer reliable.

Expected behavior

Either:

1. The string representation of Table for XLSX remains backward compatible and preserves stable row boundaries,

or

2. A documented, stable API is provided to extract structured row and column data from the Table element without relying on whitespace formatting of str(Table).

The previous behavior allowed deterministic header extraction and key value reconstruction. That is no longer possible using the same logic.

Screenshots

If helpful, attach:

• Output of str(Table) using 0.16.0
• Output of str(Table) using 0.16.1 or 0.16.10

Highlight that the row grouping changed.

Environment Info

Python version: 3.12

unstructured versions tested:
0.16.0 works as expected
0.16.1 fails to preserve row boundaries
0.16.10 exhibits same behavior as 0.16.1

OS: macOS

Please let me know if additional environment details are required.

Additional context

Downstream logic depends on reconstructing header to value mappings for simple sheets by:

1. Treating first row as headers
2. Aligning subsequent rows positionally
3. Emitting Header colon Value formatted text

This logic worked reliably until 0.16.0.

If relying on str(Table) for row separation is not supported behavior, guidance on the correct structured extraction method for XLSX tables would be appreciated.

Start with the partition_xlsx entry point and the Table element behavior, then reproduce the same XLSX output on versions 0.16.0 and 0.16.1 or later. Trace the whitespace or HTML minification changes mentioned in the issue and verify that row boundaries are either preserved compatibly or that a stable structured extraction path is available.