Feature request: allow disabling chat completions persistence (zero data retention)

[meffmadd]

🚀 Describe the new functionality needed

Allow operators to disable Chat Completions persistence by setting the inference store reference to null in a run config:

storage:
  stores:
    inference: null

Semantics when disabled:

• No InferenceStore is constructed, no inference_store table is created, and no background write workers are started — so no chat completion payload is ever persisted.
• Streaming and non-streaming Chat Completions keep working.
• The history endpoints (list, retrieve, messages) surface that persistence is not configured (HTTP 501) rather than returning empty data or a 404.

💡 Why is this needed? What if we don't build it?

We host OGX internally as a shared Chat Completions service. By default the inference store persists every request/response payload, so the database fills up and there's no way to disable it. On top of that, we need to guarantee zero data retention for our users — something always-on persistence makes impossible.

Other thoughts

• Backward compatible by default: omitting the key keeps persistence enabled; only an explicit inference: null disables it. This follows the same optional-store pattern the Responses store already uses.
• Other stores (responses, datasets, eval, files, prompts, vector_io) are unaffected.
• Is null the right way to express "disable this store", or would a boolean (e.g. inference.persist: false) be preferable? Note the responses store is already default=None, so null matches existing config semantics.
• Should disabling persistence also apply to the Responses API / conversations (openai_conversations) store?
• Would there be interest in a TTL/retention policy as an alternative to all-or-nothing opt-out?
• Anything else that still writes to inference_store that this opt-out doesn't cover (cf. Don't write to inference_store when calling Responses API ogx-ai/ogx#2846)?

Implementation exists on branch disable-inference-store (PR: #1).