fix(epub): retry a malformed XHTML section as HTML in loadDocument

[chrox]

Problem

A manifest can declare a section as application/xhtml+xml while the file is not well-formed XML. Adobe InDesign and Digital Editions ship an unclosed <meta charset="utf-8"> constantly. The XML parser then hands back a parsererror document whose body is null.

The render path already recovers from this. loadReplaced reparses as text/html when it sees a parsererror or a namespace-less root:

if (mediaType === MIME.XHTML && (doc.querySelector('parsererror')
|| !doc.documentElement?.namespaceURI)) {
    item.mediaType = MIME.HTML
    doc = new DOMParser().parseFromString(str, item.mediaType)
}

loadDocument, which backs Section.createDocument(), did not, so every off-screen consumer got the error document instead of the chapter.

Impact

Readest converts KOReader CREngine XPointers through createDocument() whenever the target section is not the rendered one, and walks document.body.children to resolve the path. Against a body-less error document that threw:

TypeError: Cannot read properties of null (reading 'children')

which silently killed cross-device reading progress sync.

Fix

Apply the same fallback in loadDocument.

Verification

Confirmed in real Chromium and WebKit (via Playwright) that for such a file doc.body === null and doc.querySelector('parsererror') matches, so the same test loadReplaced uses is the right one here. Covered by a regression test in the Readest repo (foliate-epub-malformed-xhtml.test.ts) that builds an EPUB with a malformed chapter and asserts both createDocument() and the full XPointer conversion; it reproduces the TypeError verbatim without this change.

Refs: readest/readest#5625