Bound OpenAI REST retries and requests

[Danielxu0208]

Summary

• make every REST request consume one bounded retry attempt
• add a configurable per-request timeout for direct and proxied requests
• stop logging quota-exhausted API keys
• add mock regressions for persistent 429/500 responses, timeout forwarding, and key redaction

Why

Several continue paths ran before the retry counter incremented. A service that continuously returned a non-200 response could therefore loop forever. Requests also had no timeout, and the insufficient-quota path wrote the full API key to logs.

Validation

• persistent 429 and 500 responses stop after the configured call count
• timeout reaches requests.post
• quota handling still disables invalid keys without exposing their value
• relevant pre-commit hooks passed
• modified files compile successfully