Refactor: Enhance Memory Core HealthService for Robustness and Diagnostics

The HealthService within the Memory Core has been significantly refactored to improve robustness, diagnostics, and overall user experience.

Key Improvements:

1. Intelligent Caching: Implemented a 5-minute caching layer for health checks. Only 'healthy' statuses are cached, allowing for immediate detection of recovery when a user resolves an issue (e.g., starts the ChromaDB). This reduces unnecessary load on the database.
2. Granular Status Levels: The service now reports three distinct statuses:
   • healthy: All systems are operational.
   • degraded: The core database is running, but optional features (like summarization, which requires a GEMINI_API_KEY) are unavailable.
   • unhealthy: Critical dependencies like the ChromaDB are down or collections are missing.
3. State Transition Logging: The service now logs clear, user-friendly messages when its state changes (e.g., from unhealthy to healthy), providing immediate feedback that a user's corrective actions were successful.
4. Gatekeeper Function: A new ensureHealthy() method has been added, which acts as a guard in other tool handlers. It prevents operations from executing if the system is not in a healthy state, providing clear, actionable error messages to the agent.
5. Improved Diagnostics: The health check response now includes a details array with human-readable messages and a features object indicating the status of optional functionalities.

These changes make the Memory Core more resilient and easier to debug for developers and AI agents interacting with it. The corresponding openapi.yaml has also been updated to reflect the new, more detailed health check response schema.