Parser document state still keeps only deduped chunks

[suguanYang]

Problem

Trace 019e1abc4a46b94ec4874f74d5ec13ac for staging job job_f2fcb52b2db7 showed the parser produced and exported 72 chunks, but final document/retrieval state only published 11 chunks because publication dedup reused the deduped chunk list for document_chunks.

Expected

Result ZIP, job_chunks, and document_chunks should all preserve the full parsed result chunk set. Deduplication should remove duplicate local media after export and record stats, but it should not shrink document-facing chunk state.

Evidence

• PR fix: preserve full parsed chunks in result ZIP #30 fixed result ZIP and job_chunks preservation.
• The same staging trace still logged 61/72 duplicate chunks skipped and graph publication with chunks=11.

Fix Direction

Keep publication dedup stats and local media cleanup, but publish the full normalized parsed chunk list during job finalization.