Add retry logic and rate limiting for production resilience

[trsdn]

Summary

Improve reliability with retry logic for Work IQ calls and rate limiting to prevent abuse.

Tasks

• Add exponential backoff retry wrapper for Copilot session calls
• Implement rate limiting middleware for API endpoints
• Add request timeout handling with proper cleanup
• Add circuit breaker pattern for Work IQ failures
• Add session cleanup guard (timeout after inactivity)

Acceptance Criteria

• Failed Work IQ calls retry up to 3 times with exponential backoff
• Rate limit: max 10 requests/minute per IP on /api/news-stream\
• Sessions are cleaned up after 5 minutes of inactivity
• Circuit breaker opens after 5 consecutive failures