fix(modalprocessors): preserve chunk document IDs during extraction

[lllleolin-max]

_create_entity_and_chunk initially stores the correct full_doc_id in both chunk stores, then _process_chunk_for_extraction overwrites the vector record with the chunk ID. For example, a chunk belonging to doc-original ends up pointing to chunk-... in the vector store while its text record still points to doc-original.

Preserve full_doc_id from the stored chunk during extraction. Records without that field retain the existing chunk-ID fallback.

Validation:

• Five regression cases cover the creation/extraction chain with batch mode on/off and an explicit document ID or None, plus extraction from a legacy record missing the field. Baseline: 2 failed, 3 passed; fixed: 5 passed (python -m pytest tests/test_modalprocessor_chunk_document_id.py -q).
• Full test suite: python -m pytest tests -q — 400 passed, 1 skipped.
• Repository Ruff 0.6.4 lint/format checks and git diff --check pass on the changed files.

The regression tests use recording stores and mocked extraction/merge calls; no external database or LLM service is required.

AI assistance was used to investigate, implement, and test this change.