Persist OpenAI Responses continuation history across instances

[zhefox]

Summary

• Restore previous_response_id history when OpenAI Responses requests are converted to Chat providers while preserving ordered tool-call transcripts, stable call_id values, parallel calls, and all supplied tools.
• Persist versioned, API-key-scoped continuation records through the shared RuntimeState Redis backend with a six-hour TTL, covering sync responses, native streams, and sync-to-stream bridges.
• Write streaming history before response.completed becomes client-visible to avoid immediate cross-instance continuation races.
• Add Redis multi-instance/reinitialization coverage, continuation regression tests, diagnostics, and deployment guidance for memory versus persistent Redis backends.

Validation

• cargo test -p aether-ai-formats (754 passed)
• cargo check -p aether-gateway
• cargo check -p aether-gateway --tests
• cargo test -p aether-runtime-state redis_runtime_instances_share_ttl_kv_across_reinitialization -- --nocapture against a temporary Redis 7 container
• cargo fmt --all -- --check
• git diff --check