fix(epub): safely extract metadata text without crashing on None nodeValue or nested elements

[hsusul]

Summary

• Fixes EpubConverter metadata extraction crash (AttributeError: 'NoneType' object has no attribute 'strip') when Dublin Core metadata elements in EPUB content.opf contain nested XML/HTML tags or produce a None nodeValue.
• Recursively collects text node values across child nodes instead of assuming node.firstChild.nodeValue is a non-None string.

Problem

In xml.dom.minidom, all Node objects (including Element nodes) define a nodeValue property whose value is None for element nodes. EpubConverter._get_all_texts_from_nodes checked hasattr(node.firstChild, "nodeValue"), which always evaluates to True for element children (e.g. <dc:title><span>Structured</span> Title</dc:title>). Attempting to call .strip() on None raised AttributeError and caused the entire EpubConverter.convert() process to fail.

Root cause

Line 144 of _epub_converter.py assumed node.firstChild.nodeValue was a string if hasattr(node.firstChild, "nodeValue") returned true. When node.firstChild is an Element, nodeValue is None, raising AttributeError: 'NoneType' object has no attribute 'strip'.

Fix

Added _collect_node_text to recursively collect text content across text and CDATA child nodes under any metadata element, stripping whitespace and filtering out empty metadata entries.

Testing

• pytest -k test_epub_metadata_nodevalue — PASS (1 passed)
• ruff check src/markitdown/converters/_epub_converter.py tests/test_module_misc.py — PASS (0 errors)
• mypy src/markitdown/converters/_epub_converter.py --ignore-missing-imports — PASS (Success)
• git diff --check — PASS

Compatibility

100% backward compatible. Preserves valid text extraction for simple string metadata while preventing crashes on formatted or nested metadata elements in EPUB documents.