Improve error handling and response parsing robustness

[ErnieAtLYD]

Problem

Current error handling has some edge cases and the OpenAI response parsing could be more robust, especially for malformed or unexpected responses.

Solution

Enhance error handling, add retry logic, and improve response parsing reliability.

Tasks

• Add exponential backoff retry logic for transient failures
• Improve OpenAI response parsing with better fallbacks
• Add minimum content validation before response parsing
• Implement circuit breaker pattern for API failures
• Add structured error logging with correlation IDs
• Improve error messages for different failure scenarios
• Add health check endpoint for service status

Error Handling Improvements

Retry Logic:

• Exponential backoff for 5xx errors
• Immediate retry for network timeouts
• No retry for 4xx client errors

Response Parsing:

• Validate minimum response content length
• Better handling of malformed structured responses
• Graceful degradation when parsing fails
• Default responses for parsing failures

Circuit Breaker:

• Open circuit after consecutive failures
• Half-open state for recovery testing
• Fallback responses when circuit is open

Acceptance Criteria

• Transient failures are automatically retried
• Response parsing never throws unhandled exceptions
• Circuit breaker prevents cascade failures
• Error messages provide actionable guidance to users
• All errors are properly logged with context

Priority

Medium - Improves reliability and user experience

Related

Error handling edge cases from PR #12 code review