Handle percent-encoded EPUB manifest hrefs

[xujiantop-crypto]

Summary

EPUB manifest href values are URI references, so filenames containing spaces or other escaped characters can appear as percent-encoded paths such as chapter%201.xhtml.

The current converter compares those raw href values against ZIP member names, which means a valid EPUB containing OPS/chapter 1.xhtml can silently omit the chapter content.

This PR resolves manifest hrefs relative to the OPF file and decodes the URI path before matching ZIP members.

Changes

• Resolve EPUB spine item hrefs via urljoin relative to the OPF path.
• Decode percent-encoded URI paths before looking up ZIP members.
• Add a regression test with href="chapter%201.xhtml" and ZIP member chapter 1.xhtml.

Verification

• python -m pytest tests/test_module_vectors.py::test_convert_local[test_vector14] tests/test_module_vectors.py::test_convert_stream_with_hints[test_vector14] tests/test_module_misc.py::test_epub_percent_encoded_manifest_href -q
• python -m black --check src/markitdown/converters/_epub_converter.py tests/test_module_misc.py
• git diff --check

I also ran broader vector tests locally, but this environment only has core dependencies installed. Vectors requiring optional extras such as docx, xlsx, pdf, and outlook fail with missing optional dependency errors unrelated to this EPUB change.