Fetch a document's metadata by ID.

Metadata response for a single document.

E1.5 additions (backward-compatible — new fields only added, none removed):

• progress_percent: 0-100 extraction progress for UI polling.
• updated_at: already existed; promoted to explicit field (was included).
• error_message: present when status='failed' or 'partial'. Machine reasons set by the money ceilings: cost_cap_exceeded (24-hour ceiling, the allowance comes back within a day) and cost_cap_monthly_exceeded (calendar-month ceiling of the free tier, comes back on the 1st, 00:00 UTC; ADR-083). A partial XL parent carries the heaviest code of its failed parts (monthly over daily); a part failure for any other reason leaves the older "N/M parts failed" text there. ADR-091: when the engine did not finish a document the reason is one of model_unavailable (the parse model did not answer), index_write_failed (the model answered, the index could not be written) or processing_failed (anything else, the outcome could not be confirmed). A large document whose parts ALL failed is failed (not partial) and carries the heaviest code of its parts: the two money codes first, then these three. Only the code is ever stored, never the text of an engine error.
• processed_at: set when status='indexed'.

ADR-036 §2.6 addition:

• content: decrypted plaintext returned to authenticated tenant owner. Present on GET /v1/documents/{id}; absent on list responses (content excluded from list for bandwidth — clients should fetch individual documents when they need the full text).

ADR-056 / BE-3 additions (Документы S — backward-compatible, all Optional/None):

• content_summary: First ~100 chars of doc text from lightrag_doc_status (no LLM — raw excerpt). None when doc_status absent.
• chunks_count: Number of text chunks extracted. None when unavailable.
• entity_count: Graph nodes extracted from this document. None when unavailable.
• edge_count: Graph edges extracted from this document. None when unavailable.