[codex] Add conversation memory ingest

[brianmeyer]

Summary

• add first-class conversation memory helpers plus storage indexing for root conversation memories and per-turn child memories
• expose memory_add_conversation over MCP and include memory evidence paths in search/search_batch responses
• update docs/UAT coverage for the expanded 24-tool MCP surface and conversation rollups

Validation

• python3 -m pytest -q
• bash tests/uat/test_mcp_server.sh
• python3 -m compileall -q src tests
• .venv/bin/python -m pip wheel . -w /tmp/recallforge-wheel-rec75

Notes

• .venv lacks twine and executable build module locally, so twine check/python -m build could not be run here. The wheel build succeeded through pip's PEP 517 path.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 5722c94dff

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]

Fall back to text when content is blank

Treating content as authoritative via raw_turn.get("content", raw_turn.get("text")) rejects valid turns when clients send both fields but leave content empty (for example, serializers that always include it) while populating text. In that case normalize_conversation_turns raises INVALID_INPUT and the entire memory_add_conversation call fails, even though the payload satisfies the documented "content or text" contract.

Useful? React with 👍 / 👎.