fix: restore double-newline row boundaries in Table.text (#4235)

[alvinttang]

Summary

Fixes #4235 — str(Table) no longer preserves row boundaries after 0.16.0, breaking downstream parsing that relied on str(table).split("\n\n") to reconstruct header→value mappings from XLSX sheets.

Root cause: HtmlTable.text in unstructured/common/html_table.py was using " ".join(self._table.itertext()) which walks all text nodes in document order and flattens everything into a single space-separated string. This lost all row structure.

Fix: Replace the flat itertext() join with iter_rows() / iter_cell_texts() — cells within a row are joined with " ", rows are joined with "\n\n", and all-blank rows are suppressed. The metadata.text_as_html field (compact HTML) is unchanged.

# Before (broken — all cells on one line)
"1 1 0 www.example.com 2 2 0 www.example2.com ..."

# After (row boundaries restored)
"Name Score URL\n\nAlice 1 www.example.com\n\nBob 2 www.example2.com"

Changes

• unstructured/common/html_table.py — HtmlTable.text now iterates rows and joins with "\n\n"
• test_unstructured/common/test_html_table.py — updated existing text assertion; added regression test it_separates_rows_with_double_newlines_for_boundary_reconstruction

Test plan

• All 25 existing test_html_table.py tests pass
• New regression test confirms str(table).split("\n\n") returns one entry per row
• Empty rows (all cells blank) are correctly suppressed

🤖 Generated with Claude Code