Background

The exponential backoff retry added in #63/#66 now wraps both embedGemini and embedOpenAI (resolved during the #65 merge). However the OpenAI path was not part of the original retry test coverage.

Work Required

• Add unit tests for embedOpenAI retry behaviour (mock HTTP 429/500 responses)
• Add unit tests for generateOpenAIText retry behaviour
• Verify RetryConfig is respected for OpenAI calls (same as Gemini)
• Consider whether callOpenAIJSON should be injectable for testing

Related

• feat: implement exponential backoff for Gemini API #63 (exponential backoff for Gemini)
• fix: harden retry logic - typed error checking and injectable RetryConfig #66 (typed error checking and injectable RetryConfig)
• Add OpenAI support #65 (OpenAI support)