feat: incremental extraction — cb feed for adding/updating docs

[rajnavakoti]

Description

Currently adding a new document requires re-running the full extraction pipeline. Add a cb feed <file-or-folder> command that incrementally processes new or updated documents without touching existing entities. Uses the existing source document provenance and crash-safe resume infrastructure.

Acceptance Criteria

• cb feed <file> extracts entities from a single new document and adds them to the existing KB
• cb feed <folder> processes only files not already in the KB (by filename + content hash)
• When a previously-processed file is fed again with changed content, existing entities from that file are updated (not duplicated)
• Entities from unchanged source docs are not re-processed
• Accumulated knowledge is updated with new entities after feed completes
• Output clearly shows: N new entities created, M entities updated, K entities unchanged

Out of Scope

• Structural fingerprinting for section-level diffing (P2)
• Entity deletion when content is removed from source doc (needs design)
• Cross-document concept synthesis