[codex] Add index-versioned query cache

[brianmeyer]

Summary

• Add model- and index-version-aware query cache keys for repeated text, image, video, and generated expansion inputs.
• Persist a durable storage index version in LanceDB cache metadata and bump it only for visible index mutations or batch promotion.
• Document the cache invalidation behavior and add regression coverage for version changes and hidden staged batches.

Validation

• python3 -m pytest -q
• python3 -m compileall -q src tests
• git diff --check
• bash tests/uat/test_mcp_server.sh
• .venv/bin/python -m pip wheel . -w /tmp/recallforge-wheel-rec115
• PYTHONPATH=/tmp/recallforge-twine .venv/bin/python -m twine check /tmp/recallforge-wheel-rec115/recallforge-0.3.0-py3-none-any.whl

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 42bc919878

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Derive cache model ID from active embedder model

_cache_model_id() only inspects model_name/model_id/model/_model_name, then falls back to the backend class name, but the shipped backends track the active embedding model in fields like EMBEDDER_MODEL (and MLX can change it at runtime via set_model_ids). In that case the cache key does not change when the embedder model changes, so repeated queries can reuse vectors generated by the previous model, producing retrieval in the wrong embedding space after a model switch.

Useful? React with 👍 / 👎.