Fix odd-length hex text extraction

[cnYui]

Fixes #176.

Summary

• Pad odd-length hex strings before bytes.fromhex() in string watermark extraction.
• Keep the existing errors="replace" decode behavior, so attacked/cropped images may still decode to replacement text instead of claiming recovery.
• Add a unittest regression case for the odd-length hex path and run it in GitHub Actions.

Tests

• python -m unittest discover -s tests
• python setup.py install
• python examples/example_bit.py
• python examples/example_img.py
• python examples/example_no_writing.py
• python examples/example_str.py
• python examples/example_str_multi.py
• git diff --check