Fix title extraction truncated at the first character reference - #402

HtmlRewriter feeds HTMLParser with convert_charrefs=False, so that entity and character references are re-emitted verbatim in the rewritten content. That hands the text of a <title> over in several calls, and handle_data kept only the first piece:

warc2zim uses this value as the ZIM entry title (Rewriter.rewrite_html()), so a title with an ampersand or an escaped apostrophe ends up truncated in title search and suggestions.

The pieces are now collected as they arrive, with references decoded, and joined when the title element closes. Nothing is added or changed in the output stream, so content is exactly what it was; only the extracted title changes.

This keeps both behaviours the current implementation has, which is why I went for it rather than the two BeautifulSoup variants discussed in the issue (a second parse of the document costs ~15% on the rewrite, per the measurements in #401):

• the first <title> which has text still wins: with an empty <title></title> in head, the title of a later inline SVG is used;
• whitespace around the title is still stripped, and only around it;
• a nested tag in a title still cuts it, since html_rewrite_context is per-tag and I did not touch that.

One case changes beyond the bug itself: a document whose title element is never closed (<title>Unclosed &amp; title) yielded its first piece, and now yields its whole text, so _save_title() also runs at close().

Tests cover named (é), decimal (&, '), hex (') and bare-& references, whitespace stripping, a second title in body, an empty head title followed by an inline SVG, an unclosed title, and that the references survive in the rewritten content.

ruff check / ruff format and pyright (strict) are clean, and tests/rewriting passes (635).

Fixes #401