Improve robustness of graph cycle detection code

You have written code to detect cycles in a directed dependency graph of services, where nodes represent services and edges represent dependencies between services.

In a real-world production environment, this code will be part of a larger system that validates service configurations before deployment.

Question:

How would you improve this cycle detection component so that it works better in a real-world setting? Discuss considerations and potential improvements in areas such as:

• Code readability and maintainability
• Handling large-scale graphs (performance, memory)
• Error handling and edge cases (e.g., invalid or missing data)
• Logging, monitoring, and observability
• Testing strategy (unit tests, integration tests, property-based tests)
• API design (inputs/outputs, how results are surfaced to other services or to users)
• Extensibility for future requirements (e.g., reporting all cycles, not just existence)

Provide concrete examples and rationale for the improvements you propose.