Relative links in a page whose URL contains // are resolved with the slashes collapsed #346

Follow-up to #344 / #340.

When a page's own URL contains consecutive slashes, relative links inside it are resolved with urllib.parse.urljoin, which drops the empty segments. So the // gets collapsed before normalize() even sees the URL.

Example (from @Sinkleberg's check in #344): rewriting other.html from https://example.com/x//y/page.html gives the ZIM path example.com/x/y/other.html, but a browser (WHATWG URL resolution) resolves it to https://example.com/x//y/other.html. So the link points to an entry that doesn't exist.

This is in ArticleUrlRewriter.__call__:

item_absolute_url = urljoin(
    urljoin(self.article_url.value, base_href), item_url
)

The same applies to base_href resolution. A fix probably needs an RFC 3986 / WHATWG-style join that keeps empty segments instead of urljoin. #344 already has a small resolver like that in its tests.