Add more error handling to the model training page

[tk-king]

Training can get stuck at various stages:

• Initialisation: No datasets are suitable for the current configuration
• Feature extraction: Window_size and / or step size is too large => Maximum one window could be extracted

[tk-king]

Started to implement this here: https://github.com/edge-ml/ml/tree/feat/train_error_handling