MCP: include document breadcrumb (collection + ancestor titles) in document responses

Summary Document responses across the MCP tool surface (list_documents, fetch, create_document, update_document, move_document) include collectionId and parentDocumentId as UUIDs, but no human-readable location. This is a meaningful gap for AI agents, who often see only the document's title, URL, and content — without the contextual cues a human reader gets from the sidebar. Why this matters for AI use A model reading a document benefits enormously from knowing *where* the document lives. The same article means different things under "Engineering › Onboarding" vs "Sales › Playbooks › Archive". Concretely: • Disambiguation. Wikis routinely have multiple docs titled "Overview", "Getting Started", or "FAQ". Without ancestry the agent can't tell them apart when summarizing or citing search results. • Authority and scope. A policy under "Legal › Approved" carries different weight than a draft und…