Surface error to user during fit if training data contains null values

[srinify]

Problem Description

CTGAN isn't designed to work with training data that contains missing values. However, a user can fit a CTGAN model on data with missing values and an unhelpful error is currently only surfaced during the sampling phase.

You can find code and stack traces in issues like:

• Error when sampling if original dataset contains null values #413
• ValueError: mismatch of shapes when sampling data for compas dataset #329

Expected behavior

CTGAN should surface an InvalidDataError or similar error during fit if data contains missing values:

InvalidDataError: CTGAN doesn't support null values in the training data. Please remove all null values from your training data.