Make example notebook runnable outside Kaggle

[muskan-110]

Summary
This PR improves the portability of the example notebook so it can be run outside Kaggle without errors, interactive prompts, or environment-specific assumptions.

Changes

• Removed Kaggle-specific assumptions for local execution (dataset paths and secrets)
• Made dataset paths environment-aware and added a clear error when the dataset is not available locally
• Disabled Weights & Biases by default for local runs to avoid interactive login issues
• Switched to tqdm.auto to ensure progress bars work reliably across Jupyter, VS Code, and terminal environments
• Relied on dependencies declared in requirements.txt instead of hardcoded import paths

Notes

• Kaggle behavior remains unchanged
• When running locally without the dataset, the notebook now fails early with a clear and actionable error message