feat(extract): store extraction model name on documents

[cpcloud]

Track which LLM model was used for structured extraction as a column on the documents table. This helps users know what model produced the extraction results and informs re-extraction decisions (e.g., re-extract with a better model).

Acceptance criteria

• New extraction_model column on documents
• Populated when LLM extraction completes (accept flow)
• Displayed in the Documents table
• Cleared when document has no extraction results