Support non-root entry points and nested directories in HTML5 zips

[rtibbles]

Summary

HTML5 zips were rejected unless index.html existed at the archive root, even though Studio accepts such zips on upload. This ports Studio's findFirstHtml/cleanHTML5Zip behavior to the conversion pipeline: detect the HTML entry point (index.html at the common root, then any index.html, then the shallowest HTML file), strip a common parent directory shared by all archive members, and record a non-default entry point in extra_fields.options.entry so Kolibri loads it.

Two existing test fixtures asserted the old behavior (zips with only notindex.html or a nested index.html are invalid) and were updated, since those archives are now intentionally valid.

References

Ported from Studio's frontend/shared/utils/zipFile.js. Extracted from in-progress spreadsheet chef work; no linked issue.

Reviewer guidance

uv run --group test pytest tests/pipeline/test_convert.py tests/test_files.py tests/test_data.py

Areas worth a careful look:

• _find_entry_html priority order (ricecooker/utils/pipeline/convert.py) — should match Studio's findFirstHtml exactly so ricecooker and Studio agree on the entry point
• _prepare_archive rewrites denested zips through a temp file that is cleaned up in a finally — check the error paths
• Validation loosening is intentional: archives that previously failed now pass

AI usage

I used Claude Code to extract this change from a larger working branch and add the tests; tests were verified to fail without the change, and I reviewed the final diff.