Add Presidio Evaluator to AI Evaluation Assistant

[negruber1]

This pull request introduces several improvements and fixes to the AI Assistant evaluation backend and frontend, focusing on dataset portability, evaluation metrics visualization, and dependency management. The most significant changes include making dataset paths portable across machines, enhancing the evaluation UI to display metrics per entity type, and updating dependencies for smoother setup and operation.

Dataset portability and management:

• Refactored how dataset stored_path and path are saved in the registry to be relative instead of absolute, making dataset references portable across different machines. The loader now gracefully handles legacy absolute paths and attempts to resolve them if not found.
• Updated the example dataset entry and added a new sample medical records dataset in datasets.json, using relative stored_path values. [1] [2]

Evaluation UI improvements:

• Enhanced the evaluation page to display metrics per entity type in a table, in addition to overall metrics per config. This includes new UI components and data structures for precision, recall, and F1 scores for each entity type. [1] [2] [3]
• Replaced the bar chart visualization of overall metrics with a more readable card-based layout per config.
• Removed unused imports related to the previous charting implementation.

API and dependency updates:

• Changed the evaluation API call from fetching runs to triggering a new evaluation run, aligning the frontend with backend expectations.
• Added presidio-evaluator as a backend dependency and updated setup instructions to include the required spaCy model download for local evaluation. [1] [2]

Miscellaneous fixes:

• Cleaned up package-lock.json by removing unnecessary "peer": true fields from several dependencies. [1] [2] [3] [4] [5] [6] [7] [8]

These changes improve the robustness, usability, and maintainability of the evaluation workflow for AI Assistant.

[RonShakutai]

Great PR !!,
One bug all the rest looks good
Lets take it together next week.

[RonShakutai]

there is an issue when in the ground truth there are two entities that overlap.
In this case the page is broken.

This raises a question: if the user tags the same area twice and they overlap, the page crashes. Should we limit the user in the UI, or should we resolve it in the backend?
We need to decide.

If I go back to the tagging and make sure there are no overlapping entities in the golden dataset, the error is resolved.