openai/gpt-chat-latest → A (4:1)

Side A makes a substantive storage-model redesign: it replaces the durable deque with an append-only list, removes post-commit trimming, updates the in-memory representation to `Vec`, bumps the schema version, and adds a test verifying query-time capping of recent votes. Side B mainly centralizes base-URL generation around a `HOSTNAME` helper, fixes redirect/callback URL construction, and adjusts the deployment workflow to staging, which is useful but narrower in long-term architectural impact.