fix: RSS feed metadata leakage and PPTX table pipe escaping

[JasonOA888]

Summary

Two bugs found and fixed in the converter layer.

Bug 1: RSS/Atom metadata leakage

_get_data_by_tag_name() used getElementsByTagName() which recursively searches all descendants. When a channel/feed lacked its own <title> or <description>, the method would return the first matching element from an <item>/<entry> instead, causing:

• Item titles appearing as the feed title
• Item descriptions appearing as the feed description
• Incorrect title in DocumentConverterResult

Fix: Added _get_direct_child_data() that only searches immediate child elements, and used it for channel/feed-level metadata. Also fixed UnboundLocalError when channel_title is None and channel_description is not.

Bug 2: PPTX table pipe escaping

_convert_table_to_markdown() used html.escape() on cell text, which does not escape pipe characters (|). When cell content contained |, the resulting markdown table had extra columns.

Fix: Added .replace("|", "\|") after html.escape().

Reproduction

RSS bug:

from markitdown import MarkItDown
import io

rss = b"""<?xml version="1.0"?>
<rss version="2.0"><channel>
<description>My Channel</description>
<item><title>Item Title</title><description>Item Desc</description></item>
</channel></rss>"""

md = MarkItDown()
result = md.convert_stream(io.BytesIO(rss), file_extension=".rss")
# Before: result.title == "Item Title" (wrong)
# After:  result.title is None (correct)

PPTX bug:

# PPTX with a table cell containing "|"
# Before: | Alice | Has a | pipe |  (3 columns, broken)
# After:  | Alice | Has a \| pipe |   (2 columns, correct)

Test plan

• All 112 existing tests pass (6 failures are pre-existing missing outlook optional dep)
• Manually verified both bugs with reproduction scripts