Summary

• keep same-provider credential-pool recovery for most providers
• prefer configured fallback immediately for google-gemini-cli / cloudcode-pa rate limits
• add focused regression tests for the CloudCode fallback decision

Testing

• /Users/justinkausel/Documents/GitHub/hermes-agent/.venv/bin/python -m pytest -o addopts='' tests/agent/test_gemini_fast_fallback.py -q
• /Users/justinkausel/Documents/GitHub/hermes-agent/.venv/bin/python -m py_compile run_agent.py
• VPS: /root/src/hermes-agent/venv/bin/python -m py_compile run_agent.py
• VPS: timeout 120 /root/src/hermes-agent/venv/bin/hermes chat -Q -q 'Reply exactly: GEMINI_DRIVER_OK'