Index ePub documents like PDFs

[Sriram-PR]

Fix #333

PyMuPDF, which scraperlib already uses for PDF indexing, also opens ePub. So instead of a new parser, get_pdf_index_data now goes through a shared _get_document_index_data(filetype=...), and get_epub_index_data is the ePub entry point on top of it. The title is built the same way from the document metadata (title - author). StaticItem auto-indexes application/epub+zip items in the content, fileobj and filepath cases, next to the PDF branch.

The filetype is now passed explicitly to pymupdf.open() for both, rather than relying on detection from the stream.

MuPDF says unknown epub version: 3.0 for every EPUB 3 file but reads it fine, so that message is added to IGNORED_MUPDF_MESSAGES. Real ePubs also make it warn about images it cannot load (html: cannot load image src='...') and CSS it does not support (syntax error: css syntax error: ..., e.g. on the img[src*="..."] selectors in Wikisource exports). Those carry the file name or the CSS, so they go in a new IGNORED_MUPDF_MESSAGE_PREFIXES matched on the start of the line. Neither matters for the text. Without these, practically every ePub item would log a "PyMuPDF issues" warning.

Tried on a real book first: Gutenberg's Pride and Prejudice EPUB 3 (24 MB with images) gives 733k characters of text and "Pride and Prejudice - Jane Austen" as title in about 0.3s.

Tests: a small EPUB 3 is built by a session fixture in tests/conftest.py (two chapters plus title/author metadata, a missing image and one such CSS selector), so there is no binary file to review. New tests cover get_epub_index_data from content, fileobj and filepath, check that none of the warnings above is logged, and check that an auto-indexed ePub item can be found in a ZIM by full-text search and by title suggestion. The item tests fail without the items.py change, and the warning check fails if the new ignored message is removed. Locally: tests/zim 311 passed (and the warning check fails without the new ignored messages), coverage of indexing.py and items.py is 100%, and ruff check, ruff format --check and pyright are clean. The only failures in the full suite are 22 gif tests, which fail the same way on main because I don't have gifsicle installed.

One thing worth flagging: since auto_index defaults to True, scrapers that already add ePub files will start indexing their content after upgrading, as happened for PDFs. That is what #333 asks for, but it means more build time and a bigger index for them. Scrapers that don't want it can pass auto_index=False. papers already does, so for openzim/papers#43 the plan is to call get_epub_index_data explicitly for the one format picked per book.

CHANGELOG entry added under Unreleased / Added.

[Sinkleberg]

Checked the EPUB paths with synthetic books on Python 3.14. The indexing suite passed (23 tests, one skipped), plus 12 extra checks covering custom titles, custom index data, auto_index=False, Unicode metadata and a missing author. All three input forms preserved the book bytes in the resulting ZIM; search respected the indexing options.

I didn't find a problem in those checks. I couldn't validate the whole ZIM suite in this container: download-dependent tests failed with networking disabled or wget unavailable, and a permissions test failed while running as root.