Expose progress bars for data pre-processing during model fitting

[srinify]

Problem Description

When training synthesizer models, especially when using GAN based approaches on large datasets, it can take many hours for the computation to finish and the progress bar may only kick in once in the actual model training phase. Exposing a progress bar for the data pre-processing and other key phases would significantly improve quality of life here.

I'm not sure what all 🔑 phases exist, so this likely will require some design thinking here to make it as useful as possible.

Originally suggested here: sdv-dev/CTGAN#354