Accessibility: screen readers can read the document canvas (AccessKit nodes) #488

Gap #18 in docs/gaps.md: AccessKit is enabled in eframe, but the document canvas is one opaque widget, so screen readers (NVDA, JAWS, VoiceOver, Orca) can't read or follow the document text.

Plan:

• Expose the document through AccessKit from the canvas widget: a document node with paragraphs as text nodes (role paragraph / heading with level, list items with level), tables as table/row/cell nodes, pictures with their alt text (Parity: Review › Accessibility — Alt Text pane for all objects, decorative flag, checker depth #386/Accessibility: Alt Text pane for every object, Mark as decorative, deeper checker (#386) #396), links as link nodes.
• Text selection and caret exposed (AccessKit text selection on the focused paragraph node) so screen readers announce the caret's line/word/character and selection changes; only the visible pages ± a window are exposed to keep the tree small.
• Tests that build the AccessKit tree headlessly (egui accesskit output) and check roles, heading levels, text and caret position.

I'm working on this (vangel76); PR follows.