Add opt-in 429 fallback and GPT-6 catalog support

[pkarstaedt]

Summary

• Add opt-in upstream HTTP 429 fallback via --fallback-on-429 / CHATGPT_LOCAL_FALLBACK_ON_429.
• Use a catalog-supported configurable fallback model (default gpt-reserve) with one retry only.
• Apply the behavior through the shared request path, including OpenAI Responses and Ollama-compatible requests.
• Add tests for disabled, successful, unsupported, and service-tier fallback behavior.
• Update the Codex model catalog client version to 0.160.0, invalidate stale-version caches, and add GPT-6 Astra/Sol/Luna support from issue Support GPT-6 Astra, GPT-6 Sol, and GPT-6 Luna #125.
• Update README, Docker configuration, and environment examples.

Verification

• .venv/bin/pytest -q — 22 passed.
• git diff --check — clean.
• Manual verification against the local ChatMock instance: Luna/Sol returned 429 while gpt-reserve returned 200; with fallback enabled, Luna completed successfully and logged the fallback.

Closes #125