Add figure reference navigation, backlinks, and crop UX

[enu3379]

Background

The figure extraction engine currently produces figure cards, but references in document text, embedded PDF links, caption labels, panel navigation, and manual crop editing are not connected into one coherent round-trip workflow.

Acceptance criteria

• Extracted FigureEntry data persists without overwriting manual regions
• Text references and caption labels open the corresponding figure card without moving the document
• Embedded figure links use the same panel-mediated navigation
• Figure cards jump to the detected region with the specified alignment and flash
• Mention chips return to body references and preserve the origin reference
• Manual crop updates the stored region and card preview
• Document replacement discards stale scan results
• Retry and keyboard accessibility behavior from fix: make figure previews retryable and accessible #15 remains intact
• Unit tests and unpacked-extension QA pass

AI-assisted.