Metadata Tracking and Hyperparameter Visualization for Visdom

Currently, Visdom excels at real-time visualization but lacks persistent experiment tracking and historical run filtering. To solve this, this project introduces a lightweight, self-hosted experiment management layer built natively into Visdom using SQLite and SQLAlchemy. It adds a structured metadata store, a custom key-value query engine for filtering runs, an interactive Parallel Coordinates pane for hyperparameter visualization, and Optuna integration, bringing Visdom to parity with modern ML tracking tools.