feat: Enhance retry logic to respect Cerebras API rate limit headers

[uedag5]

This pull request enhances the rate-limiting and retry logic for API requests, specifically adding support for Cerebras API rate limits and improving test coverage for these scenarios. The main changes include introducing logic to handle the x-ratelimit-reset-tokens-minute header for Cerebras, adjusting how delays are calculated based on the API provider, and adding comprehensive unit tests to verify the new behavior.

Cerebras Rate Limiting Support:

• Added a new function getCerebrasRateLimitDelayMs to extract delay values from the x-ratelimit-reset-tokens-minute header, including proper handling of floating point and invalid values.
• Updated the delay calculation logic in getDelayDurationAndStatus to use Cerebras-specific headers when the API base URL matches Cerebras, falling back to standard Retry-After headers for other providers.

Unit Test Improvements:

• Added a new test suite for Cerebras rate limiting in retry.test.ts, covering scenarios for correct delay extraction, fallback behavior, and edge cases with invalid or missing headers.

Minor Test Cleanups:

• Removed unnecessary eslint-disable comments for promise rejection assertions in retry.test.ts for improved code clarity. [1] [2] [3] [4]
• Updated import statements to include HttpError directly from retry.js for consistency.