Fix OpenAI-compatible API error handling for DeepSeek and other providers

[roomote]

This PR addresses issue #5724 by implementing comprehensive error handling for OpenAI-compatible API providers, specifically targeting the "403 Invalid response body" and "Premature close" errors experienced with DeepSeek-R1 and other providers.

Changes Made

🔧 Error Handling Improvements

• Connection Error Handling: Added specific handling for ECONNRESET, ECONNREFUSED, ETIMEDOUT, and ENOTFOUND errors with user-friendly explanations
• Premature Close Handling: Implemented targeted error handling for "Premature close" and "Invalid response body" errors
• HTTP Status Code Handling: Added comprehensive handling for HTTP status codes (401, 403, 404, 429, 500, 502, 503) with detailed explanations

🔄 Retry Logic

• Exponential Backoff: Implemented retry logic with exponential backoff (1s, 2s, 4s delays) for transient failures
• Smart Retry Logic: Only retries on transient errors (network issues, server errors), skips retries for client errors (4xx except 429)
• Jitter: Added random jitter to prevent thundering herd problems

📝 User Experience

• Actionable Error Messages: Replaced cryptic error messages with clear, actionable guidance
• Provider-Specific Messages: Error messages include provider name and context-specific troubleshooting steps
• Backward Compatibility: Maintained existing test expectations and error message formats

🧪 Testing

• All Tests Pass: Ensured all existing tests continue to pass (354 tests passed, 1 skipped)
• Error Message Compatibility: Preserved expected error message formats for Chutes, Groq, and OpenAI providers
• Type Safety: All TypeScript type checks pass

Files Modified

• src/api/providers/openai.ts - Enhanced OpenAI provider with comprehensive error handling
• src/api/providers/base-openai-compatible-provider.ts - Added error handling to base class for all OpenAI-compatible providers

Testing

cd src && npx vitest api/providers/__tests__/ --run
# ✅ Test Files: 29 passed (29)
# ✅ Tests: 354 passed | 1 skipped (355)

Impact

This fix will significantly improve the user experience when using DeepSeek-R1, Azure AI Inference Service, and other OpenAI-compatible providers by:

• Providing clear, actionable error messages instead of cryptic technical errors
• Automatically retrying transient failures to improve reliability
• Helping users understand and resolve connection issues quickly

Fixes #5724

---

Important

Improves error handling and retry logic for OpenAI-compatible API providers, enhancing user experience and reliability.

• Error Handling:
  • Added handling for ECONNRESET, ECONNREFUSED, ETIMEDOUT, ENOTFOUND errors in base-openai-compatible-provider.ts and openai.ts.
  • Implemented handling for "Premature close" and "Invalid response body" errors.
  • Added HTTP status code handling for 401, 403, 404, 429, 500, 502, 503.
• Retry Logic:
  • Introduced exponential backoff with jitter for transient errors in retryApiCall().
  • Smart retry logic to skip retries for client errors (4xx except 429).
• User Experience:
  • Replaced cryptic error messages with actionable guidance.
  • Error messages include provider name and context-specific troubleshooting steps.
  • Maintained backward compatibility with existing tests.
• Files Modified:
  • base-openai-compatible-provider.ts: Enhanced error handling and retry logic.
  • openai.ts: Improved error handling and retry logic.

^{This description was created by for 1a9e4c4. You can customize this summary. It will automatically update as commits are pushed.}

[ellipsis-dev]

The retryApiCall function implements exponential backoff with jitter well. Consider extracting this logic into a shared utility to avoid duplication across providers.

^{This comment was generated because it violated a code review rule: irule_tTqpIuNs8DV0QFGj.}

To fix the issue, the code should parse the URL and validate its host explicitly. This ensures that the check is performed on the actual host component of the URL, preventing bypasses via embedded substrings in other parts of the URL.

Steps to implement the fix:

1. Use the URL class to parse the modelUrl and extract its host.
2. Replace the substring check (modelUrl.includes(".volces.com")) with a proper host validation using endsWith(".volces.com").
3. Ensure that the validation logic is robust and handles edge cases like subdomains correctly.