Refactor: Improve MCPServer retry logic and transient error detection

[teemow]

Summary

• Retry Logic: Added background retry loop for failed MCPServers with exponential backoff
• Transient Error Detection: Extended to include HTTP 5xx errors (500-511)
• Max Concurrent Retries: Added limit of 5 concurrent retries to prevent thundering herd
• Comprehensive Tests: Added unit tests for orchestrator retry methods

Changes

Orchestrator Retry Logic

• New retryFailedMCPServers() background task that periodically checks for failed servers
• shouldAttemptRetry() checks service state and backoff expiration
• MaxConcurrentRetries = 5 prevents overwhelming upstream services
• Proper shutdown handling with WaitGroup for in-flight retries

Transient Error Detection

• HTTP 5xx status codes (500-511) detected via loop for DRY code
• Added descriptive patterns for common 5xx error messages
• Updated comment documentation to reflect 500-511 range

Code Quality Improvements (Code Review)

• Replaced time.Sleep in test with proper channel synchronization (no race conditions)
• Added compile-time interface verification for mock types (var _ Interface = (*Mock)(nil))
• Simplified HTTP 5xx pattern matching using loop (DRY principle)
• Added test for restart error handling path
• Extracted RestartGracePeriod as named constant with documentation

Documentation

• Documented the 200ms grace period in Restart() method explaining why it's needed
• Added RestartGracePeriod constant with detailed rationale

Test plan

• All existing tests pass (make test)
• All 168 BDD scenarios pass (muster test --parallel 50)
• New retry tests cover:
  • shouldAttemptRetry with various service states and backoff conditions
  • attemptReconnectFailedServers with eligible/ineligible services
  • MaxConcurrentRetries enforcement
  • Graceful shutdown of retry loop
  • Restart error handling path
• HTTP 5xx tests updated for complete 500-511 coverage
• No linting errors