add `read_only` mode for codex.query

[aditya1503]

This PR introduces asynchronous call for codex.query, while we wait for Trustworthy RAG scores, which helps reduce latency by around 35-40% for TLM Validator, we’re explicitly trying to avoid creating Codex entries for good responses, previously a read_only mode was available, this will be very useful to bring it back.