fix(xlsx): XLSX emits std minified .text_as_html

[scanny]

Summary
Eliminate historical "idiosyncracies" of table.metadata.text_as_html HTML introduced by partition_xlsx(). Produce minified .text_as_html consistent with that formed by chunking.

Additional Context

• XLSX .text_as_html is minified (no extra whitespace or thead, tbody, tfoot elements).
• table.text is clean-concatenated-text (CCT) of table.

[Coniferish]

Do we need a test case for find_subtable=True, infer_table_structure=False?

[scanny]

I'd actually be inclined not to have any tests like this :) This test boils down to "it doesn't raise for this arg combination", which is an impoverished test at best.

But I found a bug while I was in there on the html-minification and needed a test that triggered that bug so I could confirm my fix and defend against regression.

I think the broader remedy would be to break up partition_xlsx() a bit, like into a _XlsxPartitioner that we could write isolated unit tests for, but I didn't want to get into the extra scope.

Anyway, as it turns out, that particular case is handled by test_partition_xlsx_infer_table_structure() on line 95.

[Coniferish]

Sounds good to me then. I had to look up what these parameters both did and wasn't positive on their interaction. It did cross my mind to add find_subtable to the docstring, but I assumed you would be against that. The docs might need to be updated to reflect that xlsx uses infer_table_structure, though, because I did query the ai search on the site and it said "The infer_table_structure parameter is not directly applicable to XLSX files in Unstructured."

[scanny]

I think both those have a place in the docstring. Definitely find_subtable, since it's XLSX-specific. infer_table_structure too because it's used in the partitioner code.

I've added find_subtable. infer_table_structure was already there. Sounds like the AI got that wrong.

[Coniferish]

LGTM